Umts PDF

Advances in
UMTS Technolog
This page intentionally left blank
INNOUATIUE TECHNOLOGY SERIES

INFORMATION SYSTEMS AND NETWORKS
Advances in
UMTS Technology
edited by
JC Bic & E Bonek
London and Sterling, VA
First published in 2001 by Hermes Science Publications, Paris

First published in Great Britain and the United States in 2003 by Kogan Page Science, an imprint of
Kogan Page Limited
Derived from Annales des Telecommunications, Vol. 56, no. 5-6, GET, Direction Scientifique, 46 rue
Barrault, F 75634, Paris, Cedex 13, France.
www.annales-des-telecommunications.com
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the
publishers, or in the case of reprographic reproduction in accordance with the terms and licences issued
by the CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at
the undermentioned addresses:
120 Pentonville Road
London N1 9JN
UK
www.koganpagescience.com
22883 Quicksilver Drive

Sterling VA 20166-2012
USA
Hermes Science Publications and GET, 2001

Kogan Page Limited, 2003
The right of J C Bik and E Bonek to be identified as the editors of this work has been asserted by them
in accordance with the Copyright, Designs and Patents Act 1988.
ISBN 1 9039 9614 7
British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library.
Library of Congress Cataloging-in-Publication Data

UMTS, l'evolution des technologies. English
Advances in UMTS technology / edited by J. C. Bik and E. Bonek.
p. cm. -- (Innovative technology series: information systems and networks)
Includes bibliographical references and index.
ISBN 1-903996-14-7
1. Global system for mobile communications. I. Bik, J. C., 1950- II. Bonek, Ernst.
III. Title. IV. Series.
TK5103.483.U48 2003
621.3845 '6--dc21
2002040643
Typeset/Design by Jeff Carter, London

Printed and bound in Great Britain by Biddies Ltd, Guildford and King's Lynn
www. biddies.co. uk
Tontents
Foreword
J. C. Bic, E. Bonek
1. Third generation mobile systems UMTS/IMT-2000
J.-P. Charles
VII
1
2. Improvements in W-CDMA: principles and experimental results

M. Sawahashi, K. Higuchi, S. Tanaka, F. Adachi
12
3. Multicarrier CDMA techniques for future wideband wireless

networks
M. Helard, R. Le Gouable, J.-F. Helard, J.-Y. Baudais
61
4. Interpretations and performances of linear reception in downlink

TD-CDMA and multi-sensor extensions
L. Ros, G. Jourdain, M. Arndt
92
5. Smart-antenna space-time UMTS uplink processing for system

capacity enhancement
T. Neubauer, E. Bonek
126
6. Radio network planning process and methods for W-CDMA

J. Laiho, A. Wacker
146
7. An open software-radio architecture supporting advanced

3G+ systems
C. Bonnet, G. Caire, A. Enout, P. Humblet, G. Montalbano,
A. Nordio, D. Nussbaum, T. Hohne, R. Knopp, B. Rimoldi
177
8. Wireless communications + + +
R. Steele
196
Index
213
Foreword
In recent years enormous research effort has been devoted all over the world
to specify, create and develop efficient radio interfaces and access network
architectures in order to provide new services. Research laboratories, mobile
operators, manufacturers, regulators have all contributed to the definition of
a world-wide system. This so-called third generation mobile system is now
coming to reality in Europe and Japan by the name UMTS (Universal Mobile
Telecommunication System). The main features of UMTS are now well known:
Spectrum efficient radio interfaces based on spread-spectrum and CDMA
techniques, and sophisticated modulation and coding methods offering
high capacity.
Large bandwidth enabling broadband services with bit rates several times
larger than enhanced second-generation systems, even if the 2 Mbit/sec bit
rate per user would likely be limited to picocells.
Ability to interconnect with IP-based networks, paving the way to truly
fixed-mobile networks convergence.
Flexibility of mixed services with variable data rates, providing a wide
range of services from low-rate speech to interactive multi-media
communications.
Now one of the most exciting challenge for the coming years is the deployment of these complex networks both from technical and financial viewpoints.
Even if the planning is not so optimistic as it was one year ago, operations will
certainly begin in 2002.
New services are crucial for the success of UMTS. Although the general
service principles are stated (Open Service Architecture), the "killer application"
is still well kept in the drawers of operators and manufacturers, and that is why
this aspect is not deeply investigated in this publication.
In parallel with the implementation of the standards, research especially on
the air interface is still proceeding at a rapid pace for even better capacity,
quality and flexibility with enhanced transmitters/receivers.
This publication will address several issues related to UMTS emphasizing
future evolution to improve the performance of Third-Generation Wireless
Mobiles on the way to Fourth Generation. The contributions come from
academic scientists, manufacturers and operators.
VIII Foreword
The first contribution, "Third generation mobile systems UMTS/IMT-2000" by

J.-P. Charles describes the process that lead to UMTS in different standardization
bodies, ITU, ETSI, 3GPP, and gives an overview of the resulting main characteristics
for radio interfaces, network architecture and service principles.
The second chapter " Improvements in W-CDMA: Principles and Experimental
Results" by M. Sawahashi, K. Higuchi, S. Tanaka and F. Adachi, reviews
several critical aspects of the radio interface, channel code structure, spreading
code assignment, rate matching and diversity. It proposes new techniques such
as interference cancellation and adaptive antenna diversity for enhancing link
capacity. Laboratory and field trial results illustrate the improvements provided
by these techniques.
New access methods called MC-CDMA are introduced in the third chapter
"Multicarrier CDMA Techniques for Future Wideband Wireless Networks" by
M. Helard, R. Le Gouable, J.-F. Helard and J.-Y. Baudais. MC-CDMA combines
code division techniques, DS-CDMA type, and multi-carrier techniques, OFDM
type, methods. Their advantages in terms of capacity are demonstrated in the
context of an UMTS environment. MC-CDMA turns out be a promising candidate
for UMTS evolution.
The fourth chapter "Interpretations and Performances of Linear Reception in
Downlink TD-CDMA and Multi-sensor Extensions" by L. Ros, G. Jourdain and
M. Arndt focuses on modelling the multi-user TD-CDMA UMTS downlink channel
and analyses the performance of multi-user detection in various indoor and
vehicular environments, highlighting the benefits of joint detection and diversity
reception.
Performance of smart antennas is investigated in the fifth chapter "Smartantenna Space-time UMTS Uplink Processing for System Capacity
Enhancement" by T. Neubauer and E. Bonek. Space-only and space-time
processing techniques in the FDD mode with different service mix and system
loading provide enhanced capacity by a factor of 2.5 or greater, depending on
the mix of traffic services and system loading.
Deployment questions are addressed in Chapter six "Radio Network Planning
Process and Methods for W-CDMA" by J. Laiho and A. Wacker. It stresses traffic
profile and radio access technology as the most significant challenges
for system dimensioning and radio network planning for a third generation
W-CDMA system. Coverage is cell and service specific as opposed to second
generation networks. Static radio network planning simulator results are
compared to those of a dynamic simulator and are shown to be adequate for
planning purposes.
The main characteristics of a versatile real-time test platform are described in
the seventh chapter "An Open Software-radio Architecture Supporting Advanced
Foreword
IX
3G+ Systems" by C. Bonnet, G. Caire, A. Enout, P. Humblet, G. Montalbano,

A. Nordio, D. Nussbaum, T. Hohne, R. Knopp and B. Rimoldi. Such test-beds
are essential to try out and to validate new techniques proposed for the
evolution of UMTS. The platform presently implements the physical layer of the
UMTS/TDD mode, but could be extended to include new features such as multiuser detection or multiple antenna signal processing.
Finally Chapter eight is "Wireless Communications +++" by R. Steele, where
the author expresses his views on the possible evolution of wireless networks.
After recalling the recent past of second and third generation mobile systems,
new concepts such as High Altitude Platforms, body-LANs, software agents, are
discussed in the prospect of future wireless communications.
The editors would like to express their sincere thanks to all the contributors to
this book.
J. C. BIC
Ecole Nationale Superieure des Telecommunications,
France
E. BONEK
FTW, Forschungszentrum Telekommunikation Wien,

Austria
Institut fur Nachrichtentechnik und Hochfrequenztechnik
Technische Universitat Wien,
Austria
Chapter 1
Third generation mobile systems

UMTS/IMT-2000
J.-P. Charles
France Telecom R&D
I. Introduction
With third generation mobile systems, the world of mobiles will enter the era of
multimedia. The stakes are considerable: around 2010, mobile traffic should be
equal to that of fixed telephony. The convergence of mobile and Internet worlds,
the strong dynamics of innovation, and the reduction of costs in these domains
will open new opportunities for multimedia services. These systems could be
brought into service as early as October 2001 in Japan, and around 2002 in
Europe, in new frequency bands around 2 GHz.
I.1 Support of mobile multimedia services

The subscriber, at the beginning of the twenty first century, will use one or
several mobile terminals (Figure 1) for different kinds of communications: the
classical mobile phone, the pocket videophone, and the mobile PDA to manage
diary, transportation, email, and to receive multiple information. With his
portable PC, he will be connected to his company's intranet, and will benefit
from videoconference service and all facilities needed to work outside his office.
Figure 1. Mobile multimedia terminals for UMTS.
Several specific applications will use the capacities of UMTS systems to provide
data, images or even videos: video medical diagnosis, reporting, proximity
services, remote control, information, and driving guidance. Professionals' needs
will also be satisfied through access to different means of telecommunication.
UMTS will provide true mobile offices, even in vehicles.
Beyond professional use, the reduction of costs will lead to the generalization
of these personal multimedia tools, the use of which should gradually extend to a
large customer base following mobile telephony. Young people will spur the
development of this market through their needs for games, education, sports.
Thus, by the end of 2004, according to a number of studies, there will be
120 million multimedia mobiles out of a total of 1.1 billion subscribers in the
world, and 4 out of 10 Internet users will also use mobile access to Internet at that
time. Concerning data rates, UMTS is expected to offer up to 2 Mbit/sec, whereas
GSM/GPRS can only support around 100 kbit/sec.
I.2. UMTS: a global mobile system

UMTS will offer a service of universal mobility, based on the success of GSM. It
will be possible to access the same service independently of the environment:
home, office, street, car, train. It will therefore be necessary to offer a great
diversity of radio coverage schemes, from macrocells to picocells for indoor
usage. The introduction of roaming agreements between UMTS operators will
extend the geographic zone where the subscriber can access the mobile network.
As UMTS will be largely adopted by existing GSM operators, but also by others
which were not initially part of the GSM community (in Japan, for example),
UMTS subscribers will be able to use their terminals in more countries.
Although the existence of other third generation systems will limit the ability
to roam among the different systems, the fact that UMTS has been developed to
ensure backward compatibility with GSM will be a key factor for the future,
allowing a smooth transition between these two systems.
I.3. Migration from GSM to UMTS
The progressive migration of GSM networks towards UMTS appears to be essential
to preserve the considerable investments already made in second generation
mobile systems and to minimize the cost of introducing UMTS. To spread out the
investments, UMTS will be deployed at the beginning in "islands" and GSM will
ensure the continuity of service on the whole territory, but with limited services
(voice and low data rates). This scenario is based on the existence of dual-mode
GSM/UMTS terminals when the first UMTS networks are launched in Europe. For
existing GSM operators, an upgrade of their core network will be possible since
the UMTS core network is an evolution of GSM/GPRS, but they will have to deploy
Advances in UMTS technology
a completely new radio access network. Several thousand UMTS base stations
will be needed to offer national coverage with high data rates and reuse of
existing GSM radio sites will be a key issue to deploy rapidly.
II. International research and standardization

context
II. 1. Main players
In Europe, the development of a new mobile system has been largely based on
research programmes launched by the European Commission in the early 90's.
Japan followed another direction: most 3G developments were financed by
mobile operator NTT DoCoMo. Japanese industry supported this R&D effort i
order to develop a new standard and take the lead in this very competitive
market. European manufacturers (Nokia, Ericsson) took part in this effort which
led to the establishment of a common solution between them and Japan. A
compromise was reached when ETSI (http://www.etsi.org) was looking for
candidates for its third generation mobile system (UMTS). As in the United States,
a large part of the frequency band allocated by WARC 92 (World Administrative
Radio Conference 1992) for the IMT-2000 systems is currently used by second
generation systems (PCS personal communication systems); it is thus not
surprising to note that the American proposals for IMT-2000 often correspond to
evolutions of existing second generation systems in order to maintain backward
compatibility with them.
In this context, the standardization activities led within the regional (ETSI for
Europe, TTC and ARIB for Japan, TIA and ANSI for the United States) and
international organizations (ITU-R and ITU-T) developed with increasingly close
contacts, as a certain convergence among the proposals took shape (in particular
between Europe and Japan). In Europe, it is necessary to highlight the strong
position of lobbies in third generation standardization (GSM Association, UMTS
Forum) which are striving to federate, as far as possible, the stances of GSM
operators and manufacturers. The regulation authorities play, in the same way, a
fundamental role for the use of the spectrum identified by the WARC 92 and for
the attribution of UMTS licences.
II.2. The standardization of third generation mobile systems

II.2.1. Standardization in ITU
The standardization of third generation mobile systems emerged in the ITU with
the ambition of defining a global standard which would replace the existing
systems. There could be no global mobile system without a common spectrum

for all regions, therefore work on third generation systems really started once the
WARC 92 had identified new frequency bands for IMT-2000 (Figure 2). This
system, initially called FPLMTS (future public land mobile telecommunications
system), then IMT-2000 (International mobile telecommunications) was expected
to be launched at the beginning of 2000 using all or part of the spectrum
identified around the 2 GHz band. This system was expected to offer high data
rates, multimedia services, and global roaming.
Today, standardization harmonization on IMT-2000 is conducted in ITU-R/WP
8F (http://www.itu.int/imt) for the radio interface and in a new commission
recently created in ITU-T for the signalling and networks aspects. The borders
between the two entities still remain fuzzy for the protocols of the radio
interface, taking into account the distribution of the activities between the two
Figure 2. IMT-2000 spectrum.
sectors of ITU: standardization in ITU-T, Radiocommunications in ITU-R. It should

be noted that the ITU development sector (ITU-D) is highly interested in IMT-2000
because many developing countries are waiting for such a technology to provide
access to high data rate services with limited infrastructure.
In November 1996, the ITU-R approved the selection methods for the IMT2000 radio interface. A call for candidates was then launched in March 1997,
with June 1998 as the deadline to submit proposals for the IMT-2000 radio interface.
Some technical evaluations were given at the end of September 1998. However,
the ITU-R could not establish a consensus on any one of these proposals and, as a
result, five different solutions were adopted in November 1999 :
CDMA 2000 (evolution of the American CDMA IS-95 solution originally
developed by Qualcomm);
UMTS/W-CDMA (one of the UMTS modes supported by NTT DoCoMo, Nokia
and Ericsson, and developed by the 3GPP);
UMTS/TD-CDMA: second UMTS mode supported by Siemens. This mode is
also developed by the 3GPP and it also includes a specific option developed
for China;
uwc-136 (evolution of the American solution ANSI-136 or D-AMPS); this

solution integrates an evolution of GSM called EDGE (Enhanced Data rates
for GSM Evolution);
DECT developed by ETSI.
It is primarily the existence of different second generation systems (GSM, IS95, D-AMPS) which prevented a greater convergence between these various
solutions. The operators wished to preserve at least a part of the investments
already made in the infrastructures while ensuring progressive migration
towards the third generation.
II.2.2. Standardization in ETSI and 3GPP
In 1991, ETSI created technical sub-committee SMG5, to develop a third
generation mobile system called UMTS (Universal Mobile Telecommunication
System). This sub-committee was part of the technical committee SMG in charge
of standardizing GSM in order to facilitate the migration of GSM towards UMTS.
During the first years, this sub-committee co-ordinated the European positions
for the ITU meetings. When ETSI had decided to propose a solution for IMT-2000,
it became necessary to adopt a more flexible organisation to better define the
European solution which would be proposed. Therefore, standardization activity
on UMTS was distributed throughout the existing GSM technical sub-committees.
The first stage of the standardization process for UMTS was to define technical
requirements for the radio interface, mainly based on the work done in ITU-R,
and the selection process. This process was launched at an ETSI conference in
December 1996, during which various solutions were presented. Among these
proposals, three solutions prepared by the European project acts/frames were
presented (France Telecom R&D was part of this project).
France Telecom R&D was one of the rare participants to compare technically
the various solutions in competition. After a vote, during an extraordinary
meeting of the SMG technical committee in January 1998, a compromise was
found based on two harmonized modes: W-CDMA [1] and TD-CDMA [1, 4]. W-CDMA
was adopted for the FDD mode (Frequency Domain Duplex, i.e., one frequency
per transmission direction) and TD-CDMA for the TDD mode (Time Domain
Duplex, i.e., time-division multiplexing of the two directions on the same
frequency). This mixed solution offers the advantage of allowing a complete use
of the frequency bands allocated to IMT-2000: the FDD mode being used in
priority in the paired bands and TDD mode in unpaired bands. This compromise
was then submitted to ITU-R as the European proposal for the radio interface of
IMT-2000.
The adoption by ETSI of FDD/W-CDMA opened the doors for an agreement with
Japan, about to adopt this technology for its own third generation mobile system.
Discussions among various standardization organizations: ETSI for Europe, TTC
and ARIB for Japan, TTA for Korea, T1 for the United States, led to the creation, in
December 1998, of a partnership among these organizations called 3GPP (third
generation partnership project). This forum (http://www.3gpp.org) developed the
technical specifications for UMTS. Then, these specifications were adopted as
standards by the different national or regional standardization bodies.
III. Radio Interface

III.l. Objectives
Some of the objectives and constraints were defined before the design of the
UMTS radio interface. These objectives strongly influenced the choice of the
parameters of the various proposals, and it is necessary to point them out.
The UMTS radio interface was built to support a broad range of different
services, with higher data rates than those offered by second generation systems
(GSM, IS-95, PDC,...) (see Table I). UMTS offers circuit switched or packet switched
mode services, with a maximum data rate depending on the environment and the
speed of the mobile. Services with variable and asymmetrical data rates (between
uplink and downlink) will be supported in an efficient way. Table I gives some
performances: binary error rate (BER), delays for different types of services.
UMTS will be deployed in a multilayer cellular network, with macrocells (0.5
to 10 km) for overall coverage, microcells (50 to 500 m) for hot spots, and
picocells (5 to 50 m) for indoor coverage. Handover will be ensured in a
transparent way for the user, without any perceptible cut or degradation of
quality.
UMTS will use spectral resources in an efficient way, by adapting the
protection of the transmitted data to the radio channel. It will be necessary to
optimize capacity and coverage. At the beginning, coverage will be the main
goal of UMTS operators and then, gradually as the traffic increases, it will be
necessary to increase capacity.
Planning of UMTS networks will be carried out if possible using automatic
procedures. However, as for CDMA, coverage and capacity are closely linked,
operators will need to use suitable radio planning tools in order to guarantee their
customers the radio coverage, quality of service and data rate they expect.
The need for coexistence with second generation systems, and in particular
with GSM in Europe, represents an additional constraint for UMTS. For that, it will
be necessary to provide dual-mode GSM/UMTS terminals when UMTS networks are
launched in Europe. Those terminals will be able to support handover between
GSM and UMTS, which will allow progressive deployment of UMTS.
Table I. Performance requirements for UMTS.

Environment
Rural
(v 500
km/h)
Urban
(v 120
km/h)
Indoor and
microcells
(v 10 km/h)
Non-real time services

Real time services
Max bit rate Delay/BER Max bit rate Delay/BER
144 kbit/sec
144 kbit/sec
delay
delay
20 - 300 ms
384 kbit/sec 150 ms in
BER
384 kbit/sec
95 % of
10-3 - 10-7
the cases
BER
2 Mbit/sec
2 Mbit/sec
10-5 - 10-8
III.2. The radio interface chosen by ETSI and developed by 3GPP

As indicated above, the solution adopted by ETSI in January 1998 is based on
two harmonized modes: FDD/W-CDMA [1] for the paired bands and TDD/TD-CDMA
[1, 4] for the unpaired bands.
In the compromise adopted by ETSI, it was also stated that the UMTS system
could be deployed using only 2 x 5 MHz band, and that the selected parameters
would ensure harmonization with GSM and dual-mode operation FDD/TDD while
maintaining the objective of a low-cost terminal.
FDD mode is appropriate for all types of cells, including large cells, but is not
well adapted to support asymmetrical traffic. TDD is by definition more flexible to
support traffic asymmetry, but it requires synchronization of the base stations, and
is not appropriate for the large cells due to the limited guard periods between time
slots. Table II gives the main characteristics of the two UMTS modes.
FDD mode is based on CDMA with a wide bandwidth (5 MHz). One of the
major differences with IS95, developed by Qualcomm in the early 90s, is that
no synchronization is needed between base stations, thus allowing easier
deployment for operators. One of the key advantages of CDMA is its high spectral
efficiency, so that UMTS operators will be able to offer, with the same spectrum,
higher data rates than with GSM. When offering the same services as for GSM
(voice for example), CDMA will give them more capacity per MHZ: recent
evaluations have shown that the gain in terms of spectral efficiency could be in
the order of 2 or 3 [5, 6].
TDD mode is based on a mix between TDMA and CDMA. Basically, the TDD frame
has 15 time slots and, for each time slot, there is a possibility to support several
simultaneous CDMA communications when joint detection is used.
Table II. Main characteristics of TDD and FDD modes.

Mode
Multiple access
Bit rate
Carrier spacing
Frame length
Frame structure
Modulation
Spreading factor
Channel coding
FDD
TDD
(Frequency domain
duplex)
(Time domain duplex)
DS-CDMA
TDMA/CDMA
3.84 Mchip/s
4.4 to 5 MHz with a 200 kHz raster
10 ms
15 time slots per frame
QPSK
4 to 256
1 to 16
Convolutional (rate 1/2 to 1/3)
Turbo codes for BER < 10-3
IV. Network infrastructures

IV.I. General architecture
Figure 3 presents the general architecture of the UMTS network. It shows that
UMTS is not only one new radio interface, but also a complete mobile network
based on an evolution of the GSM/GPRS core network.
The UMTS core network comprises two distinct domains: circuit switched (CS)
and packet switched (PS), as in GSM/GPRS networks. The core network's elements
are the same: MSC (Mobile switching centre) for CS services and SGSN and GGSN
for PS services. Two solutions are available to introduce UMTS: either to upgrade the
existing elements or to introduce new ones supporting UMTS.
The principle of the separation between the access network and the core
network through a standardized interface remains as in GSM. This new interface
(Iu) is a reference point which, according to the different implementations, may
correspond to one physical interface or two. However, there are always two
distinct logical flows through this interface: one for the packet switched domain
and the other for the circuit switched domain.
The concept of the subscriber identification module (SIM) is kept for UMTS, but
with a new smart card: theUICC(UMTS integrated circuit card). This card supports a
GSM SIM for GSM subscribers, the USIM for the UMTS subscribers, as well as other
modules for different applications (credit cards, e-commerce, subscriptions for
leisure activities).
IV.2. Access network architecture

Figure 4 represents the logical architecture of the UMTS access network. The
radio network subsystem (RNS) includes the radio base stations (node B), and
their controller (RNC).
Figure 3. General architecture of the UMTS (release 99).
This hierarchical architecture, in which an entity controls several entities at a

lower level, is similar to that of the GSM radio access network (BSC-BTS). Iu
represents the interface between the RNC (Radio network controller) and the core
network. Iub represents the interface between the nodes B and the RNC. The main
difference with GSM is the existence of the Iur interface between RNCS. The main
reason for the introduction of this interface is the management of macrodiversity
(soft handover mechanism) in the access network. This interface will enable the
management of soft handover between two node B's belonging to two separate
RNCS, independently from the core network.
ATM was chosen for transport in the access network. This choice makes it
possible to support all types of services (voice, circuit data, packet data, ...) that
will be offered. Different AALS (ATM Adaptation Layers) will be used: AAL2 for
the user data (voice or data) on the interfaces Iu-cs (circuit switched domain), Iur
and Iub. AAL5 is used for signalling and the user data on the Iu-ps interface
(packed-switched domain).
Figure 4. UMTS radio access network architecture (UTRAN).
10
V. Service principles in UMTS

V.1 Open Service Architecture (OSA)
For GSM, the different services were fully standardized: voice, fax, short
messages, supplementary services (call hold, call forward, call conference, ...)
but, for the operators, it was difficult to propose innovative services to attract the
customer. So, in order to provide greater flexibility in service creation, it was
decided during the second phase of GSM standardization to introduce "toolkits":
CAMEL (concept of intelligent network for GSM), SIM toolkit, and MexE (Mobil
Execution Environment), which includes WAP (Wireless Application Protocol
These toolkits were used in GSM to introduce prepaid services (CAMEL) or mobil
internet portals (WAP). For UMTS, these principles are still valid but efforts
are focused on integrating all these toolkits in a single one called OSA (Open
Service Architecture). OSA is, in fact, an API based on PARLAY (PARLAY
(http://www.parlay.org) is a forum developing a common API for the differen
networks). This new concept is still under development in 3GPP and will be
introduced in the next UMTS releases.
V.2 Virtual Home Environment (VHE)
The VHE concept will enable the customer to use his services with the same
ergonomics independently of his location; thus it will be possible to provide him
with the same environment in his home network and when he is roaming. CAMEL
(Customized Applications of Mobile network Enhanced Logic), originally
developed for GSM networks, will provide roamers with the same services the
use when they are in their home network.
CAMEL is based on an intelligent network architecture which separates service
logic and data base from the basic switching functions, and implements the CA
(CAMEL Application Protocol) derived from INAP (Intelligent Network
Application Protocol). When a subscriber is roaming, all his CAMEL data, which
are stored in the HLR (Home Location Register), are transferred to the visited
network. Thanks to this mechanism the service provided has the same
ergonomics wherever the subscriber is.
VI. Conclusion
The choice of the principles of the UMTS radio interface in January 1998 gave a
strong acceleration to the standardization process throughout the world. This
decision was particularly important, because it consolidated the technical
agreement between Japan and Europe on the adoption of CDMA as a common
11
basis for UMTS. However, this was only the first step leading to the launch of
UMTS networks in October 2001 in Japan and in 2002 for Europe. In 2000, most
of the European countries have allocated UMTS licences using beauty contest or
auctions procedures, to give sufficient time to the UMTS operators to prepare the
launch of their services in 2002. A first release of the UMTS standard which is
called release 99, was adopted at the beginning of 2000, and this release will be
used by the manufacturers for the first generation of UMTS equipment. The
competition between operators will mainly be based on their ability to provide to
their customers new services because, when UMTS is launched, a high percentage
of the population will have a mobile for telephony and it will be very difficult,
especially for a new entrant, to attract new customers with existing services. The
key aspect of UMTS will be access to high data rates and multimedia services for
the customer and, without such services, it will be difficult to transform this
costly adventure into success.
REFERENCES
[1] HOLMA (H.), TOSKALA (A.), wcDMA for UMTS, John Wiley & Sons, (2000).
[2] MOULY (M.), PAUTET (M-B.), The GSM system for Mobile Communications, (1992).
[3] BLANC (P.), CHARBONNIER (A.), VERRIER (D.), L'UMTS: la generation des mobiles
multimdia, L.'cho des recherches, n 170, 1er trimestre, (1998).
[4] HAARDT (M.), KLEIN (A.), KOELHN (R.), OESTREICH (S.), PURAT (M.), SOMMER (V.),
ULRICH (T.), The TD-CDMA based UTRA TDD mode, IEEE Journal on Selected Areas
in Communications, 18, n 8, pp. 1375-1384, (Aug. 2000).
[5] Acx (A.G.), MENDRIBIL (P.), Capacity evaluation of the UTRA FDD and TDD modes,
49th Vehicular Technology Conference, Houston, 3, pp. 1999-2003, (1999).
[6] FRANCE TELECOM, Technical analysis and comparison of UTRA concepts, ETSI SMG2
Adhoc n 4, Tdoc SMG2/UMTS 126/97, Helsinki, (17-21 Nov. 1997).
Chapter 2
Improvements in W-CDMA:
principles and experimental
results
M. Sawahashi, K. Higuchi, and S. Tanaka
Wireless Research Laboratories, Japan
F. Adachi
Graduate School of Engineering, Tohoku University, Japan
I. Introduction
Associated with the successful planned introduction of global commercial
wideband code division multiple access (W-CDMA) [1], [2] service from this year,
the dawn of the genuine era of wireless Internet is upon us. The achievable
maximum information bit rate guaranteed by the required quality level in the
IMT-2000 is 2 Mbps and in the near future the peak bit rate of nearly 10 Mbps
will be possible for high-speed downlink packet access (HSDPA), which is now
undergoing standardization in the Third Generation Partnership Project (3GPP).
Therefore, rich services such as Internet access and the transmission of video
and high-quality images from/to moving vehicles will be achieved in the wCDMA system. DS-CDMA wireless access, on which W-CDMA is based, has
numerous advantages over TDMA or FDMA including single frequency reuse, soft
hand-off (or site diversity), enhanced radio transmission through Rake
combining, and direct capacity increase through sectored antennas. The key
features of the W-CDMA physical layer are:
- Inter-cell asynchronous operation and three-step fast cell search
- Flexible realization of various levels of quality of service (QoS) for various
transport channels by rate matching associated with channel coding
- Signal-to-interference power ratio (SIR)-based fast transmit power control
(TPC) for satisfying the required quality level for a physical channel with
minimum transmit power
13
- Significant gains in link capacity and coverage through the use of many
diversity techniques, e.g., coherent Rake time diversity using pilot symbol
assisted (PSA) channel estimation, space diversity, inter-cell (sector)
diversity, and transmit diversity (only in the forward link)
- High flexibility in offering different multirate services (up to 2 Mbps)
through orthogonal variable spreading factor (OVSF) multiplexing and
orthogonal multicode transmission
- Capacity enhanced techniques such as interference cancellation (IC) and
adaptive antenna array diversity (AAAD).
The above essential W-CDMA technologies associated with its performance and
the features of the W-CDMA air-interface were comprehensively overviewed in [13]. However, in the ongoing worldwide standardization process in the 3GPP, the
radio link parameters and channel structure have been modified, and enhanced
techniques such as turbo coding for high-rate data transmission and transmit
diversity were adopted into the standards. Therefore, this chapter overviews the wCDMA enhanced wireless access technologies including the channel structure and
spreading code assignment in the physical layer and transport channel multiplexing
into a physical channel associated with rate matching and reports on a series of
laboratory and field experiments conducted in an area near Tokyo. We designed and
developed an experimental system comprising a coherent multistage interference
canceller (COMSIC), coherent adaptive antenna array diversity (CAAAD) receiver in
the reverse link, and adaptive antenna array transmit diversity (AAA-TD) in the
forward link in order to demonstrate the suppression effect on multiple access
interference (MAI) and multipath interference (MPI). The experimental results of
these techniques are also presented.
II. Physical channel and spreading code assignment

II. 1. Physical channel [4-5]
W-CDMA has a three-layered channel structure: physical, transport, and logical.
The physical channels provide several transport channels to the MAC (Medium
Access Control) layer, which is a sub-layer of the data link layer (Layer 2). The
MAC layer provides several different logical channels to a higher layer, that is the
RLC (Radio Link Control) layer. The physical channels are classified by
spreading codes, carrier frequency, and in-phase (I)/quadrature-phase (Q)
assignment.
One radio frame of a physical channel has a frame length of 10 msec and
comprises 15 slots. Thus, the slot length is equal to a basic updating unit of adaptive
fast TPC and channel estimation of coherent Rake combining and is optimized to the
14
Improvements in W-CDMA
value of 0.667 msec taking into account a tradeoff between frame efficiency and
tracking ability of fast TPC and channel estimation against fast fading variation. The
number of channel-coded information bits, which each physical channel conveys,
differs according to the type of physical channel and spreading factor (SF). The
features of the major physical channels are described below.
(1) P-CCPCH (Primary-Common Control Physical Channel)
One P-CCPCH is defined for each sector in the forward link. The P-CCPCH has
a fixed SF of 256 (15 ksps) and carries the BCH transport channel. It is not
transmitted during the first 256-chip duration, but instead the P-SCH and S-SCH are
transmitted during that period at each slot.
(2) S-CCPCH (Secondary-Common Control Physical Channel)
Multiple S-CCPCHS, which are common channels in the forward link, are
defined in each cell (sector) and carry paging information and lower data
information from a higher layer.
(3) PRACH (Physical Random Access Channel)
Multiple PRACHS, which are common channels in the reverse link, are defined
and used to carry the RACH transport channel comprising lower information data
from a higher layer.
(4) DPCH (Dedicated Physical Channel)
A DPCH is assigned to each mobile station (MS) in both the forward and
reverse links. It comprises a DPCCH (Dedicated Physical Control Channel) and a
DPDCH (Dedicated Physical Data Channel).
A DPDCH consists of a channel-coded data sequence and more than one DPDCH
can be assigned to one DPCH. A DPCCH is used for Layer 1 control of DPCH and
one DPCCH is defined for one DPCH. A DPCCH comprises pilot bits for coherent
channel estimation, TPC bits, TFCI (Transport Format Combination Indicator)
bits, and FBI (Feedback information) bits designating the control information for
transmit diversity in the forward link (thus, FBI bits are defined only in the
reverse link).
(5) CPICH (Common Pilot Channel)
A CPICH is the common pilot channel used for channel estimation, path search
for Rake combining (generation of power delay profile), and the third step, i.e.,
scrambling code identification in the three-step cell search method. Two kinds of
CPICHS are defined: primary-CPICH and secondary CPICH. The primary-CPICH has
two-symbol data sequences associated with two antennas. Without transmit
diversity all symbol sequences with all "1" are transmitted from Antenna #1, and
with transmit diversity, the second primary-CPICH with different symbol
sequences from those of the first primary-CPICH are also transmitted from Antenna
#2 in addition to the first primary-CPICH.
15
In future applications of smart antennas for spot beam transmission, the

secondary-CPICH will be defined, which will be spread by the primary or
secondary scrambling code.
(6) SCH (Synchronization Channel)
The SCH is a common channel in the forward link, which is used for cell
search. Primary and secondary-SCHS are used for the first step and second step
for the three-step cell search method. They are transmitted only during the 256chip period at the beginning of each slot.
(7) AICH (Acquisition Indication Channel)
The AICH is a common channel in the forward link used for random access
control. It is used as a pair comprising a PRACH and PCPCH.
(8) PICH (Page Indication Channel)
The PICH is a common channel in the forward link and is associated with
S-CCPCH, in which the PCH transport channel is mapped.
(9) PDSCH (Physical Down Link Shared Channel)
The PDSCH is a common channel in the forward link, which carries the DSCH
transport channel and is used for high rate packet data transmission.
(10) PCPCH (Physical Common Packet Channel)
The PCPCH is a common channel in the reverse link, which carries the CPCH
transport channel and is used for high rate packet data transmission.
Figure 1. Frame structure of DPCH (a) reverse link (b) forward link.
16
The frame structure of the DPCH in the reverse and forward links is
illustrated in Figures l(a) and l(b), respectively. The DPDCH and DPCCH are codemultiplexed into I and Q channels, respectively, in the reverse link. Since the
DPCCH with a fixed rate (SF) and DPDCH with variable date transmission are
separated from each other in the orthogonal phase, fluctuation of the amplitude
during variable transmission can be decreased. Meanwhile, the DPCCH and DPDH
are alternatively time-multiplexed within a slot in the forward link.
Table I. Spreading code assignment.ent
Forward link
Channelization code
Scrambling code
Repetition period = Data symbol period
Repetition period = 10 msec frame
User identification (4-512 chips)
Cell (Sector) identification (38,400 chips)
CPICH
#0
SF = 256
P-CCPCH
#l
SF = 256
Primary
S-CCPCH
Arbitrary
SF = 4-256
Primary (Secondary)
Primary
DPCH
Arbitrary
SF = 4-256
Primary (Secondary)
AICH
Arbitrary
SF = 256
Primary (Secondary)
PICH
Arbitrary
SF = 256
Reverse link
| DPCH
Code-channel identification in
multicode transmission (4-256 chips)
Arbitrary
SF = 4-256
Primary (Secondary)
User identification (38,400 chips)
Primary (Secondary)
II.2 Spreading code assignment [6]

W-CDMA adopts a two-layered spreading code assignment, which combines a
channelization code with the repetition period of the corresponding symbol rate
and a scrambling code with the repetition of the frame interval. The OVSF code
is used as the channelization code. The spreading code assignment for each
physical channel is given in Table I. The SF of 4 to 256 is used for S-CCPCH and
DPCH.
II.2.1. Channelization code

Starting from Cch,1,0 (1) (SF = 1), the OVSF code which has a length of 2k-1-chip
at the k-th layer, is recursively generated based on the formula given below,
resulting in the tree-structured code generation as shown in Figure 2 [7].
Figure 2. Generation method of OVSF codes.
17
The k OVSF codes of the k-th layer are orthogonal to each other. Furthermore,
any two codes belonging to different layers are orthogonal except for when one
code is not the mother code of the other. For example, Cch,2,0 and Cch,4,2 are
orthogonal to each other. When Cch,2,0 is already assigned, any code below this
code on the code tree cannot be used, this is a restriction of the code assignment.
The codes of Cch,256,0 and Cch,256,1 are commonly used for all cells for the
P-CPICH and P-CCPCH in the forward link, respectively. The channelization codes
of other physical channels are assigned from a higher layer.
II.2.2. Scrambling code

Cell (sector)-specific and user-specific scrambling codes are assigned in the
forward and reverse links, respectively. In the reverse link, the repetition period
of the scrambling code is 10 msec and that with the repetition period of 256
chips is optionally defined for future application of multiuser detection. The long
scrambling code is truncated by a duration of 38,400 chips from the beginning
of the Gold sequence with the repetition period of 224 chips. There are 224 long
scrambling codes.
The scrambling code in the forward link is generated by truncating the 38,400
chips from the beginning of the Gold sequence with the repetition period of 218 and
its shifted version by 131,072 chips. The 8,192 scrambling codes are grouped into
512 scrambling-code groups, where each group comprises 1 primary scrambling
code with 15 corresponding secondary scrambling codes. The primary scrambling
code is first used, and then the secondary scrambling codes are used to cover any
shortage in the channelization code set associated with the primary scrambling code.
Five hundred twelve primary scrambling codes are divided into 64 primaryscrambling-code groups (hereafter we simply denote group), each including 8
18
primary scrambling codes. This group-wise divided primary scrambling code

structure is used for the three-step cell search algorithm, which is described in
Section III.
II.2.3. Synchronization code

A synchronization code is used to spread a SCH and comprises a primary
synchronization code (PSC) and secondary synchronization code (ssc) both with
the length of 256 chips, which are used for P-SCH and S-SCH, respectively. Let PSC
be denoted as Cpsc, in which Cpsc is a complex-value code sequence with the
same sequence for real and imaginary parts expressed as
where
Let 16 sscs be denoted as Cssc,k (k = 1, 2, ..., 16). Then, Cssc,k is generated

by multiplying the j-th component (1 j
256) of vector Z of a common
sequence with the length 256 chips and the j-th component of the n-th column
of H8 of the Hadamard matrix, where n = 16 X (k 1). Let hn(j) and z(j) be the
j-th symbol of n-th column of the Hadamard matrix and the j-th symbol of a
common sequence, respectively. By selecting 16 columns from 256 columns
every 16 columns, the 16 Cssc,k is generated as
where
II.2.4. Spreading
In the reverse kink, the channelization code is independently spread into I/Q
channels by using different OVSH codes and weighted by weighting factor G,
which denotes the transmitted amplitude (power) ratio of DPCCH to DPDCH.
Complex spreading is applied to the physical channel: one is a code truncated by
38,400 chips from the beginning of the Gold sequence with the repetition period
of 224, and the other is truncated by 38,400 chips of the shifted first Gold
sequence by 16,777,233 chips. Thus, the spreading using channelization codes
and the scrambling codes are expressed as
19
where DI(Q) denotes the I/Q components of the chip data sequence spread by
channelization codes and CI(Q) represents the I/Q components of a long
scrambling code. In this QPSK spreading, the carrier phase transition by
-degrees occurs across the zero point, thus incurring increasing nonlinear
distortion of the power amplifier. Therefore, in the 3GPP standard, the HPSK
(hybrid PSK) scheme was adopted, which decreased the possibility of the phase
transmission crossing the zero point [6, 51]. The long scrambling codes
sequence used for spreading are generated from the two original scrambling
codes based on the following equation:
In the forward link, P-SCH and S-SCH are spread by only primary and
secondary synchronization codes, respectively, commonly used for both I/Q
channels. The other physical channels except for SCH are first spread by an
identical channelization code with SF = m for both the I/Q channels and then
complex-scrambled by the two scrambling code sequences.
III. Transport channel multiplexing

III.l. Explanation of data format for layer 1 [8]
We first explain the terminology used for data transfer between the MAC layer
and Layer 1. A transport block, which corresponds to a RLC (Radio Link Control)PDU (Protocol Data Unit), is a basic unit for data transfer between the MAC layer
and Layer 1. Cyclic redundancy check (CRC) for error detection in Layer 1 is
added to every transport block. One example of a transport block transfer
between the MAC layer and Layer 1 is illustrated in Figure 3. A set of transport
blocks simultaneously transferred between the MAC layer and Layer 1 on the
same transport channel is called a transport block set. The size of the transport
block is the length of the transport block defined in bit form. The size of each
transport block belonging to one transport block set is uniform and is a fixed
value. The number of bits within a transport block set is called the transport
block set size. As shown in Figure 3, the arrival time interval of transport block
sets between the MAC layer and Layer 1 is called the transmission time interval
(TTI), which is equal to the channel interleaving length. The TTI is some integer
20
Figure 3. Example of exchange of data between a MAC layer and Layer 1.
times the radio frame length (= 10 msec) and is defined as 10, 20, 40, or 80 msec
in the 3GPP. The transport format is a format in which a transport clock set is
transferred between the MAC layer and Layer 1 on a transport channel every TTI.
The transport format comprises two attributes: the dynamic part and semi-static
part. Attributes of the dynamic part are the transport block size, transport block
set size, and TTI, and those for the semi-static part are error of the correction
scheme such as the type of error correction and coding rate and the size of the
CRC. The transport format set (TFS hereafter) is defined as a set of transport
formats used for the transport channels. Within one TFS, the semi-static parts of
all transport channels are identical; however, the dynamic parts may be changed
every TTI in order to achieve variable rate transmission. The transport channels
are simultaneously multiplexed into Layer 1 as a coded composite transport
channel (CCTrCH). Each transport channel in the CCTrCH has an available TFS;
however, only one transport format is used at each TTI. Thus, the combination of
possible transport formats of all transport channels transferred on the same
Layer 1 at each TTI is defined as a transport format combination (TFC).
Furthermore, a set of TFC applied to the CCTrCH is called as transport format
combination set (TFCS). The indicator designating the TFC I called the transport
format combination indicator (TFCI). TFCI bits are multiplexed into the DPCCH of
each DPCH. In the receiver, the TFCI bits are used to decode Layer 1 data
sequences and de-multiplex transport blocks transferred on one physical
channel. In addition to the explicit TFCI detection method, the blind transport
format detection method using CRC to trace the surviving trellis path ending at
the zero state among the possible transport formats is also specified in the 3GPP
standard (note that blind detection is used only for the forward link) [9].
21
III.2. Transport channel [4, 8]

A transport channel is defined as a channel that is used to transfer various kinds
of data to the MAC layer. The major transport channels are described below. The
mapping relationships between the major physical channels and transport
channels are given in Figure 4.
(1) BCH (Broadcast Channel)
The BCH is a forward link transport channel that is used for broadcasting
system - and cell-specific information. The BCH is always transmitted over the
entire cell and has a single transport format.
(2) FACH (Forward Access Channel)
The FACH is a forward link transport channel that is commonly used for
multiple MSS and for transmitting low-rate user information from a higher layer.
(3) PCH (Paging Channel)
The PCH is a forward link transport channel that is transmitted over the entire
cell and is used to transmit paging information.
(4) RACH (Random Access Channel)
The RACH is a reverse link transport channel, which is received from the
entire cell. The RACH is characterized by collision risk and by being transmitted
using open-loop transmit power control.
(5) DCH (Dedicated Channel)
The DCH is a forward link and reverse link transport channel, which is
transmitted over the entire cell or only a part of the cell using a smart antenna.
The DPCH is used for the transmission of user data and is assigned to each MS.
Variable rate transmission and fast transmit power control (TPC) are applied to
the DPCH.
(6) DSCH (Down Link Shared Channel)
The DSCH is a forward link transport channel shared by several MSS. The DSCH
is used for mainly high-rate packet data transmission and is transmitted over the
entire cell or over only a part of the cell using beam-forming antennas.
(7) CPCH (Common Packet Channel)
Figure 4. Relation between physical channel and transport channel.
22
The CPCH is a reverse link transport channel and is associated with a

dedicated channel on the forward link, which provides power control and CPCH
control commands. The CPCH is used for high-rate data transmission on random
access channels.
III.3. Multiplexing and rate matching [9]
The flow of the transport channel multiplexed into a physical channel in the
reverse link is depicted in Figure 5. First, CRC parity bits required for block error
detection at the receiver are calculated for the original data sequence per
transport block of each transport channel. Then, the calculated CRC bits are
attached to each transport block. All transport blocks with CRC bits within one
TIT are serially concatenated followed by channel coding. For channel coding,
convolutional coding or turbo coding are used in the 3GPP specification. For the
common transport channels such as BCH, PCH, and RACH, convolutional coding
with the rate of 1/2 and the constraint length of 9 bits is used. Convolutional
coding with the rate 1/3 (1/2) is also used for FACH and DPCH with a lower
channel bit rate, and turbo coding [10] with the rate 1/3 and the constraint length
of 4 bits is used for FACH and DPCH with higher channel bit rates. After the coded
data sequence of each transport channel is interleaved over the length of the TTI
Figure 5. Transport channel multiplexing (reverse link).
23
(first interleaving), rate matching is performed according to the required QoS

and the number of bits. The data sequence of each transport channel after
rate matching is segmented and interleaved over one radio frame length
(second interleaving). Finally, the CCTrCH containing all transport channels is
multiplexed into a physical channel. As described previously, the first channel
interleaving is performed before rate matching of each transport channel in the
reverse link. Meanwhile, discontinuous transmission (DTX) is allowed when
there is no transmitted data sequence in the radio frame of a certain transport
channel in the forward link. Thus, the rate matching is performed independently
for each transport channel before the first interleaving.
As shown in Figure 6, transport channels with different bit rates and QoS
levels are multiplexed and transferred into one physical channel. A transport
block is a basic unit for data transfer between the MAC layer and Layer 1 (in
Figure 6 of the transport channel, 1(1) represents the first block of transport
channel 1). The required QoS, i.e., the block error rate (BLER) or bit error rate
(BER) of the physical channel is achieved by changing the transmit power or data
modulation scheme according to the fading variation. In general, the QoS level
of one physical channel can be controlled by changing the target SIR of fast TPC
using outer loop control so that the output BLER or BER is equal to the required
value as explained later. However, the average received signal energy per bit-tointerference and background noise spectrum density (Eb/Io), thus, the received
signal power, is an almost constant value during one radio frame interval.
Therefore, in order to bundle various transport channels with different QoSs into
one physical channel, the required QoSs of various transport channels are
simultaneously satisfied with respect to the identical average received signal
power by changing the number of coded bits of each transport channel after
channel decoding (this process is called rate matching). That is to say, by
repeatedly transmitting some coded bits at a regular interval, the BLER or BER is
Figure 6. Principle of rate matching.
24
Improvements in
W-CDMA
improved. Contrarily, if encoded bit sequences are punctured at a regular

interval, the received quality is degraded. In this way, the number of bits of each
transport channel multiplexed into the physical channel is flexibly changed
every radio frame by rate matching described hereafter.
In the reverse link, rate matching is performed for the coded data sequence
of each transport channel after the first interleaving. The number of bits of each
transport channel to be repeated or punctured is calculated based on the ratematching attributes signaled from a higher layer. The DTX, when there is no
coded transmitted data sequence of a certain transport channel multiplexed into
a physical channel, is not permitted. Thus, the spreading factor (SF), i.e.,
equivalently the symbol date rate, of a physical channel is first determined
according to the total number of bits per radio frame of all transport channels
multiplexed into the physical channel. Then, rate matching is performed so that
the sum of the bits of all transport channels per radio frame after rate matching
should equal the bits per radio frame accommodated into the physical channel
having the assigned SF. Let N i j and ANij be the number of coded data bits of
transport channel i per radio frame with TFC j before rate matching and the
number of bits per radio frame to be bit-repeated or punctured (the positive and
negative values of A denote the bit-repetition and puncture), respectively. The
value of Zij which is needed for the calculation of ANij is recursively computed
from the following equations using the rate-matching attribute value, RM;.
where Ndatajis the total number of bits per radio frame to be assigned to code
the composite transport channel with TFC j and 1x1 denotes the integer value
defined as x - 1 s 1x1 5 x. Using the value of Zij recursively calculated from
Equation (6), ANij is derived from the following equation.
In the reverse link, rate matching is performed per radio frame based on
Equation (7). Meanwhile, in the forward link dissimilarly to the reverse link, DTX
is applied when there are no transmitted coded data bits of a certain transport
channel. Thus, the rate-matching pattern does not necessarily change for each
radio frame. Rate matching is performed as follows. The number of bits per TI
25
of transport channel i before rate matching, NTTIi,h, is first calculated for the
corresponding TFC h belonging to TFCS. Then, from the value of NTTIi,h, and th
number of radio frames of transport channel i over TTI, Fi, the corresponding
number of bits per radio frame was derived for all TFC belonging to TFCS. Thus
rate matching is performed such that the number of total bits per radio frame for
TFC hMax, when the summation of bits per radio frame of all transport channels
is maximized, is equal to the number of bits per radio frame accommodated into
a physical channel, that is to say, the number of bits per radio frame. Then the
number of bits per TTI to be bit-repeated or punctured is computed for each
transport channel. Based on this obtained rate matching pattern, the number of
bits per radio frame of each transport channel is updated every TTI.
Consequently, when transport channels having different TTI are multiplexed, the
number of total bits belonging to a radio frame is changed at the shortest TTI at
every TTI. If the number of bits per radio frame of transport channel i after rate
matching is lower than the maximum number of bits assigned to that transport
channel, DTX is performed during an interval corresponding to the number of bits
to be shortened.
IV. Asynchronous cell sites and three-step search

method
In asynchronous cell site operation, which is the most prominent feature in
W-CDMA, flexible system deployment from outdoors to indoors is possible, since
no external timing source such as the global positioning system (GPS) is required.
To allow asynchronous cell site operation, two-layer spreading code allocation
is used [1]. In the forward link, cell sites are distinguished by their unique
scrambling codes, and data channels (control and traffic channels) in each cell
site are distinguished by different OVSF codes. To reduce the cell search time in
asynchronous cell site operation, we proposed a three-step cell search method
using scrambling code masking [11]. Subsequently, our original cell search
method was refined in the standardization process. The forward link frame
structure in the 3GPP standard required for the three-step cell search is illustrated
in Figure 7. The base station (BS) transmits a continuous common pilot channel
(CPICH), primary synchronization channel (SCH), and secondary-SCH over the
256-chip duration at the beginning of each slot (every 0.667 msec). The
spreading codes for the CPICH and the DPCHS are taken from a set of OVSF codes,
thereby maintaining mutual orthogonality between the CPICH and DPCHS. These
channels are further scrambled by a cell-specific scrambling code with a 10msec repetition period (= 38,400-chip duration), which is equal to the data frame
26
Figure 7. Forward link frame structure of CPICH and SCH.
length. The PSC for the primary-SCH is common to all cell sites and the ssc for
the secondary-sch denotes the group index into each of which the scrambling
codes are grouped beforehand. The total number of scrambling codes to be
searched is 512, which is divided into 64 groups of 8 codes each. The transmit
powers of the primary- and secondary-sch are set to half that of the CPICH.
The operational flow of the three-step cell search algorithm is illustrated in
Figure 8. Using SCHS and CPICH, the three-step cell search is performed as
follows. First, the PSC-matched filter (MF) is used. The MF output is averaged over
period T1 to detect the primary-sch time position that provides the maximum
average correlation. Next, the scrambling code group is identified by taking the
Figure 8.
Operational flow of
three-step cell
search method.
27
cross-correlation between the received signal and the set of sscs over period T2.
Finally, the scrambling code is identified by taking a partial correlation between
the received signal and each of the candidate scrambling codes and then
averaging over period T3. The scrambling code that provides the maximum
correlation is determined as the scrambling code to be searched. To reduce false
detection, a verification mode is added by using a frame synchronization check.
When the synchronization verification failed two consecutive times, the cell
search process is restarted from the first step. The correlation peaks of PSC and
ssc calculated in the first and second steps are averaged during T1 and T2 in
order to reduce the influence of MAI and the background noise components.
However, especially when the velocity of a MS is low, the probability for false
detection in the first and second steps is greater since the duration of low
received signal power due to fading becomes longer. Thus, time space transmit
diversity (TSTD) is applied to sc in the 3GPP specification, with which primaryand secondary-sc are alternatively transmitted slot-by-slot from different
antennas [5]. Since a successive primary- and secondary-SCH are transmitted
from different antennas having a low fading correlation, the false detection is
decreased due to the transmit diversity effect.
Figure 9 shows the measured laboratory experimental results of the
probability distribution of the cell search time with the fading maximum
Doppler frequency, fD, as a parameter using the 4.096 Mcps WCDMA
experimental system with TSTD [12, 13]. In addition to CPICH and schs, 10 DPCHS
without fast TPC were transmitted as a channel load. An L = 2 path Rayleigh
Figure 9. Probability of distribution of cell search time using TSTD.
28
fading channel with average equal power was assumed because we confirmed
that field experimental results conducted near Tokyo could be well approximated
using this model where 2 - 3 paths with unequal average received signal power
were observed. The transmit power ratio of CPICH to DPCH and average received
Eb/No of DPCH were set to - 3 dB and 7 dB, respectively. We set T1, T2, and T3
to 40, 30, and 10 msec, respectively. Figure 9 shows that as fD becomes larger,
the cell search time becomes shorter since false detection is decreased. The
figure also shows that by using TSTD, the cell search time when fD is low such as
5 and 20 Hz can be decreased because false detection is mitigated when the
received signal level drops. As a result, the cell search time at the detection
probability of 90% with TSTD is decreased by approximately 100 msec compared
to that without TSTD. The cell search can be completed within approximately 250
msec at the probability of 90% with TSTD, when R = - 3 dB and fD = 5 Hz.
V. SIR measurement-based fast TPC

Fast TPC based on SIR measurement of Rake combined signals is used to
minimize always the transmit power according to the traffic load both in the
reverse and forward links. This results in increased capacity by reducing the
interference to other users in other cells and the user's own-cell. Fast TPC
comprises two loops as shown in Figure 10: the inner loop and outer loop.
Inner loop operation is performed as follows. In the Rake combiner, the
despread signals associated with resolved paths are multiplied by the complex
conjugate of their channel gain estimates and summed. Therefore, if the SIR
measurement is done after Rake combining, it is affected by the channel
estimation error. In this paper, instead of measuring the SIR after Rake
combining, we apply the SIR measurement method proposed in [14, 15], in
which, first, the SIR on each resolved path is measured and then, the SIRS of all
the resolved paths are summed to obtain the SIR (which is equivalent to the one
at the output of the Rake combiner). By doing so, obtaining an SIR measurement
that has less influence on the channel estimation error is possible. The SIR
measurement is summarized below. First, signal power S l (k) of the k-th slot
associated with the l-th path is computed using the received Np pilot symbols.
Signal power S l (k) is given by
where
29
Figure 10. SIR-based adaptive TPC with outer loop control.

since we assume that the modulation phase of Np pilot symbols is /4 radians.
The instantaneous interference plus background noise power of the l-th path,
/ l (k), is computed as the squared error of the received Np pilot samples
Then, Il(k) is averaged using a first order filter with forgetting factor (< 1 )
to obtain
The SIR at the k-th slot associated with the l-th path l(k) is given by
Finally, the SIR at the k-th slot, (k), is obtained as
The measured SIR was compared to the target SIR and the TPC command w
generated, which was transmitted to raise or lower the mobile transmit power by
1 dB every 0.667 msec. Even if the received SIRS are the same, the received
quality (BLER) is not the same because the BLERS are affected by the number of
paths, maximum Doppler frequency (which depends on the speed of the
vehicle), and SIR measurement, etc. Therefore, the outer loop controls the targe
SIR with a more gradual updating interval compared to the inner loop so that the
measured BLER or BER is equal to the target value. In general, a BLER-based outer
30
loop is used. BLER is measured by calculating the number of CRC results that
coincide with the value attached to every transport block. Since the required
BLER becomes a very small value for high-speed and high-quality data
-6
transmission, e.g., with the required BER of 10 , it takes a much longer time to
calculate the BLER. As a result, outer loop control cannot track changes in the
propagation conditions. Therefore, in these cases, outer loop control based on
BER measurement of the tentative decision data symbols before channel
decoding (i.e., after Rake combining) with decision data symbols after channel
decoding as a reference can be applied. The reference data symbols are
generated by re-encoding and interleaving binary decision data symbols after
channel decoding. Although data decision error occurs in the decoded data
sequence, it is considered that the impact on the reference symbols is very small.
VI. Diversity
VI. 1. Coherent Rake combining
(Rake time diversity)
PSA coherent detection is used for both the reverse and forward links [16, 17].
The block diagram of the PSA coherent Rake combiner is illustrated in Figure
11 (a). The received multipath signals are despread by the MF and resolved into
L-multipath components of transmitted quaternary phase shift keyed (QPSK)
modulated data that are received via different propagation paths with different
delay times. The coherent Rake combiner output is expressed at the n-th symbol
position of the k-th slot associated with the l-th path (/ = 0,1, ..., L -1) using
despread signal r l (n, k), as
where l(k) represents the channel estimates. The output data sequence, d (n, k),
is de-interleaved and channel decoded to recover the transmitted binary data
sequence. In order to achieve accurate channel estimation that works
satisfactorily in a fast fading environment, we presented an improved channel
estimation filter called a weighted multislot averaging (WMSA) channel
estimation filter [17] as shown in Figure ll(b). After obtaining the instantaneous
channel estimates of each slot, the channel estimates, l(j + i)s, of 2J-multiple
slots (i = - J + 1,..., 0, 1, ..., J) are then weighted and summed to obtain the final
channel estimate, l(k), as
31
Receiv
spread
Figure 11. Coherent Rake receiver. ( a )Receiver structure (6) WWSA channel estirnationfrlter,
where ai is the real-valued weight. Using the WMSA channel estimation filter,
accurate channel estimation is possible, particularly in slow fading
environments. The optimum value of ai varies according to the fading
correlation between succeeding slots in a real fading channel. Therefore, we
proposed in [18] an adaptive WMSA channel estimation filter, in which a
weighting factor is adaptively controlled by measuring an inner product of the
averaged despread pilot signals of successive slots.
We evaluated the BER performance of coherent Rake combining with SIR
based fast TPC in field experiments conducted in an area near Tokyo. The cell site
and mobile transmitterh-eceiver antennas were located 59 and 2.9 m off the
ground, respectively. A measurement vehicle equipped with the mobile receiver
was driven along roads at distances of 0.75 - 1.35 km from the cell site at the
average speed of approximately 30 k d h . The measurement course passes
through a business zone, lined with office buildings and factories. Other
conditions are given in detail in [19]. The average delay spread of the test course
was approximately 1 psec. The test course first experienced clear two-path and
32
single-path fading at the middle of the course. Then, three-path fading with
unequal average power was observed at the end of the course.
Figure 12 plots the measured average BER performance of the 32-kbps data
rate user in the single-user and two-user cases (one-interfering user with a 64kbps data rate assuming the same BER independently employing fast TPC), as a
function of the TPC target Eb/I0 value (note Eb/I0 is calculated as Eb/I0 = SIR
101og(3/2) dB, since convolutional coding with the rate of 1/3 and QPSK data
modulation were used in the experiments) [19]. Two MSS established radi
links with BS 1. A WMSA channel estimation filter with J = 2 was used.
Laboratory experimental results of the single-user case using the L-path model
with fD = 80 Hz are also plotted for comparison. The results clearly show that
the target Eb/I0 when the interfering user exists becomes almost the same in
order to achieve the same BER as that of the single-user case, implying that fast
TPC worked satisfactorily in a real fading channel. The measured numbers of
active Rake fingers per antenna along test courses is 2.0. Figure 12 shows that
the measured BER performance is almost the same as the laboratory-measured
BER performance when L = 2. The field-measured BER performance results are
in good agreement with those estimated from the laboratory experiment. The
figure also shows that two-branch space diversity (antenna diversity) reception
can reduce the target Eb/I0 by approximately 3 dB at the average BER of 10-3.
Figure 12. Average BER as a function of target Eb/I0 per antenna.

Field experiments.
33
With space diversity reception, the average BER of 10-3 can be achieved at the
required Eb/I0 of approximately 3 dB per antenna.
The measured average BER performance of the 64 kbps channel using turbo
coding is plotted in Figure 13 as a function of the MS relative transmit power
with the channel interleaving length of TCHL = 40 msec [20]. Turbo coding with
the rate of R = 1/3 and the constraint length of K = 4 bits (generator polynomials
are 13, 15, and 15 in octal notation) were used, while the rate and constraint of
convolutional coding as a reference were R = 1/3 and K = 9 bits, respectively.
Primary interleaving (PIL) [9, 22] and multistage interleaving (MIL) [9, 21],
which offer a greater capability for randomization compared to the block
interleaving method, were used as turbo interleaving and channel interleaving
methods, respectively. In the experimental system, Max-log - Map decoding
was used as the soft-in/soft-out decoder and the number of iterations, m, was
assumed to be eight, which was sufficiently large. From Figure 13, the MS
average transmit power for the average BER of 10-6 using turbo coding can be
decreased by approximately 0.6 (0.3) dB compared to that using convolutional
coding without (with) antenna diversity reception. Although the superiority of
turbo coding to convolutional coding was confirmed in an actual multipath
fading channel, this difference was decreased compared to the laboratory
experiments assuming a fixed delay time for each path using a fading simulator,
Figure 13. Average BER of 64-kbps data transmission with turbo coding
as a junction of mobile transmit power. Field experiments.
34
i.e., superiority was confirmed to be above 1.0 dB. This abatement in the
improvement with antenna diversity reception indicated that in an actual fading
channel in the field experiments, the impact of path search for Rake combing and
SIR measurement for fast TPC diminished the improvement in performance of the
turbo coding due to a very low received signal power level.
VI.2. Site diversity (soft/softer handover)
Soft handoff or site diversity ("site diversity" hereafter) [23, 24], which was first
implemented in the IS-95 CDMA standard [25], is an essential technique together
with fast TPC in improving transmission impairment due to multipath fadin
and shadowing near the cell edge. The simplified configuration of site diversity
is illustrated in Figure 14. In the forward link, the same original information
sequences before channel coding are transferred to NBSS(N is the number of BSs
with which the MS is associated) through the back haul (wired transmission lin
between BS and radio network controller (RNC)) from a RNC and transmittedfrom
two BSS using different scrambling codes. The received signals after Rake
combining at the MS are combined symbol-by-symbol with maximal ratio
combining (MRC) followed by soft-decision Viterbi decoding. With inter-sector
diversity in the reverse link, the Rake-combined signal of each sector is
combined with MRC in the same way as in the forward link. On the other hand
with inter-cell diversity, in the reverse link, a hard-decision data sequence after
soft-decision Viterbi decoding at each BS is transferred to the RNC via the back
haul with the reliability information associated with each traffic channel. The
transferred data sequences are selection-combined every selection period,
according to the reliability information.
Figure 14. Simplified configuration of site diversity.
35
VI.2.1 Reverse link

The performance of reverse link inter-cell site diversity depends on the type of
reliability information that is used. Therefore, we present a two-step SC scheme
using two types of reliability information [26, 27]: CRC results calculated ove
selection interval TSEL and the average received SIR measured over interleavin
interval TILV. In our scheme, we use the number of slots with a measured SIR value
greater than the target value of fast TPC, NSIR, i.e., the number of TPC command bits
to lower the transmit power during TILV, instead of the actual measured SIR value
This is because the transfer capacity in the back haul required for the reliability
information of inter-cell site diversity can be significantly decreased (note that
only 4 bits/frame are required for denoting the SIR average over one frame). Th
SC at the RNC was performed in two steps.
(1) Step 1: When multiple decoded data sequences transferred from N cell
sites (BSs) indicate no CRC error, then the one data sequence over TSEL amon
the data sequences yielding the successful CRC result is selected.
(2) Step 2: When all the CRC results transferred from N BSS indicate frae
error, the data sequence during TSEL with the larger NSIR over TILV is selected.
The field experiments using inter-cell site diversity were conducted in an
area near Tokyo in order to measure the BER performance in a 32-kbps data rate
channel. The measurement course is a road running north and south, which
passes through the middle of 2 BSS. The distance between BS 1 and BS 2 i
approximately 2.5 km. The middle point of the measurement course is
approximately 1,300 and 1,200 m apart from BS 1 and BS 2, respectively. On
either side of the measurement course is a low-rise factory area. The view from
BS 1 was line-of-sight (LOS) except at the end of the course, while it was nonline-of-sight (NLOS) from BS 2 due to the tall buildings. We set the soft-handove
threshold to 3 dB. The difference in the measured average received signal
powers from the two BSS was approximately 1 dB. Thus, the measurement
course is a softhandover area within the prescribed threshold. The power delay
profile with 1 (2) and 2(1) paths were observed in the first half and the latter half
of the course from BS 1 (BS 2). We set TSEL = 10 msec. In the experiments, fast
TPC was used only in the reverse link. The received signal power was set to be
sufficiently high so that there was no TPC command bit error. The measured tim
variations of the instantaneous BER and received E b /I 0 at BS 1 and BS 2, after
inter-cell site diversity, and the mobile transmit power averaged over one radio
frame length (=10 msec) are plotted in Figure 15. The target Eb/I0 at each BS was
set to 7 dB so that the average BER after intercell site diversity was
approximately 10-3. The figures show that bit errors occurred when the received
Eb/I0 at each BS dropped; however, the instantaneous received E b /I 0 after intercell site diversity was maintained at almost a constant level. Therefore, the
36
Figure 15.
Instantaneous time variations in the reverse link inter-cell site diversity.
Field experiments.
measured BER after inter-cell site diversity significantly improved; nevertheless

the BER measured at each BS was significantly degraded due to the reduced signal
level caused by shadowing and fading variations. Since the target Eb/I0 was set
to satisfy the average BER of 10-3 after selection combining, burst error rarely
occurred since convolutional coding was used.
VI.2.2 Forward link
When fast TPC is applied in the forward link inter-cell site diversity mode, each
BS independently follows the TPC command bit sent from the MS via the reverse
link. Therefore, the transmit power of each BS differs when a TPC command bit
error occurs in the reverse link. An increase in the difference between the
37
transmit powers of the BSS causes a reduction in site diversity gain and an
increase in the interference to other users. To overcome this problem, several
schemes that compensate for the BS transmit power were proposed [28, 29]. I
the method proposed in [28], each BS controls its instantaneous transmit pow
by using a forgetting factor so that the difference between instantaneous transmit
power and the BS-specific reference transmit power calculated by averaging the
instantaneous values does not become large. However, it is difficult to quickly
track variations in path loss including shadowing due to the movement of the MS
The method in [29] reduces TPC bit error by sending the same TPC bit ov
several slots in the site diversity mode, this prevents the transmit power
difference between BSS from becoming large. However, in addition to the
problem described in [29], the TPC delay increases. Therefore, we proposed th
following two step algorithm to reduce the impact of TPC errors and keep the
transmit power of the BSS the same as that shown in Figure 16 [26].
(1) First loop: the standard transmit powers, P(k)REF, of all BSk are
compensated by P(k) according to the dedicated control channel from a MS
based on the average SIR measurement at a MS.
where the Measured_total_Eb/I0 and Target_ Eb/I0 are the measured Eb/I0 after
Rake combining and the target Eb/I0 at a MS, respectively. The P(k)REF is constan
during the length of G-slot and its value of n (= gxG)-th slot P(k)REF(n) is updated
every G-slot as P(k)REF(g x G) = P(k)REF((g-1) x G) +
Figure 16. Combination of forward link site diversity and TPC.
38
(k)
(2) Second loop: the instantaneous transmit power, P CL(n), is controlled

according to the TPC command bits (TPC) by introducing the forgetting facto
using the standard transmit power compensated in the first loop.
The measured average BER performance in the forward link when inter-cell sit
diversity is applied is plotted in Figure 17 as a function of the total BS average
transmit power. The measurement course and experimental conditions were
the same as those in Figure 15. Fast TPC was used for the reverse and forwar
links. The number of maximum Rake fingers for the BS and MS was 4. The
forgetting factor was set to R = 0.8. In the measurement course, it was
observed that the instantaneous transmit power is controlled around the
standard transmit power without dispersing to the maximum output during the
course. The figure shows that the total transmit power of the 2 BSS at the
average BER of 10-3 in the inter-cell site diversity is decreased by
approximately 0.3 dB compared to a one cell-site connection. This
improvement is small compared to that in the reverse link because the increase
in interference due to transmissions from two BSS diminished the diversity
effect. Thus, the site selection diversity transmit power control (SSDT) [30], in
which only the primary BS transmits control bits to decrease the interference,
was proposed.
Relative total average transmit power (dB)

Figure 17. Average BER in the forward link inter-cell diversity.
Fields experiments.
39
VI.3. Transmit diversity

Transmit diversity employing several antennas at a BS can improve the forward
link transmission performance without increasing the complexity of the MS
[31, 32]. Therefore, several transmit diversity schemes were adopted in the
3GPP standardization [5, 33]. The CPICHS with the same spreading code, but
different data modulation patterns, are transmitted from two antennas in the
same carrier phase. Two open-loop type transmit diversity schemes were
adopted in the3GPPstandardization: TSTD [5] and space-time transmit diversity
(STTD) [5, 34]. STTD, which is used for the CCPCH, transmits two data sequences
in parallel after coding from two antennas using different channelization
codes. Since the fading correlation between the two antennas is low, the
fluctuation in the received signal level due to fading is mitigated. The
operational principle and coding scheme of STTD is illustrated in Figure 18(a)
and Figure 18(b), respectively. Let S(m) be the QPSK symbol data sequence
denoted as S(m) = exp j(m), where (m){h/2 + /4 ; h = 0 - 3} is the QPSK
modulation phase. Then, two successive symbols, S(m) and S(m + 1) are
treated as a pair, where m denotes an even number. The two symbol sequences,
dl(m) and d2(m), for antennas #1 and #2 generated in the STTD encoder are
expressed respectively as
It is clear that the orthogonality between the two data sequences is maintained
irrespective of the spreading code sequence.
Figure 18. Operational principe of STTD.

(a) Block diagram of transmitter, (b) example of STTD encoding.
40
On the other hand, closed-loop type (two modes were standardized) transmit
diversity is used for DPCHS, in which the transmit antenna weights are controlled
by the FBI generated at the MS [33]. Let W1 = A1ej1 and W2 = A2ej2 be the
transmit antenna weights. Thus, in Mode 1, the transmitted phase of the second
antenna, 2, is changed with the accuracy of /4 according to the FBI from the M
so that the received SIR after combining is maximum. This is expressed as 1 = 0,
2 = {+/- /4, +/- 3/4}, A1 = A2 = v 1/2. Meanwhile, th transmitted
amplitudes of two data sequences are also controlled by FBI bits as well as th
transmitted carrier phase in Mode 2.
The measured average BER performance with STTD is plotted in Figure 1
when fast TPC was not applied in the forward link as a function of the averag
received Eb/I0, where I0 is the multipath interference plus background noise
power density [35]. The measurement course was course #1 described in [19].
The performance with and without antenna diversity reception at a MS is show
in the figures. The BER performance with signal antenna transmission is als
depicted for comparison. Figure 19 shows that the average required received
Eb/I0 at the average BER 10-3 with STTD was decreased by approximately 1.5
(1.0) dB without (with) antenna diversity reception. The improvement using
STTD with antenna diversity reception became smaller than that without antenna
diversity because the degradation of the channel estimation due to a lower
Figure 19. Average BER performance infield experiment

(without forward link fast TPC).
41
received level offset the additional diversity effect by STTD when using Rake
path diversity and antenna diversity reception. From the figure, the effectiveness
of STTD for a channel without TPC such as a common control channel was
elucidated in a real multipath-fading channel.
VII. W-CDMA capacity enhanced technologies

In DS-CDMA systems, due to multipath fading and shadowing as well as distancedependent path loss, severe MAI is often produced, which significantly reduces
the link capacity. In the forward link, although the orthogonality among the same
propagation channels is achieved by using OVSF channelization codes, the MPI
especially from high rates users is severe. Thus, the IC or multiuser detection
(MUD) [36, 37] and adaptive antenna array receiver [38, 39] are promising
techniques to reduce MAI and thereby decrease the transmit power of MSS in the
reverse link. AAA-TD [40-43] is effective in decreasing severe MPI in the forward
link without changing the air interface and adding complexity to the MS. High
capacity BS transceiver configuration examples are shown in Figure 20. Figure
20(a) shows the configuration using CAAAD [44, 45] and AAA-TD in the reverse
and forward links, respectively. Figure 20(b) is a configuration including IC and
AAAD. In the reverse link, the COMSIC [46-48] and CAAAD both with multiple
antenna configurations of greater than two are installed in parallel. Only the
COMSIC with multiple antenna reception is used for data demodulation in the
reverse link, while the CAAAD receiver is used only to generate the transmit
antenna weights for AAA-TD, which are created by performing both RF circuitry
calibration and carrier frequency calibration on the receiver antenna weights.
However, by using the parallel structure of COMSIC and CAAAD, two advantages
are obtained. The first is, since the interference replicas of other users for each
array antenna are independently generated in COMSIC, high-quality decoded data
symbols at the COMSIC output are used to update the transmitter (receiver)
weights in addition to pilot symbols. The second is that SIR measurement for fast
TPC associated with outer-loop control and accurate path search is possible at the
CAAAD output after MAI and MPI suppression.
VII.1. COMSIC
VII.1.1. COMSIC receiver
The COMSIC is categorized into two types: the serial (successive)-type and the
parallel-type. The parallel-type COMSIC is more practical than the serial-type due
to its inherent advantage of a short processing delay, which is derived from the
42
Figure 20. High-capacity BS transceiver configuration examples.
despreading, PSA-channel estimation, Rake combing, and generation of

interference replicas for all active users simultaneously performed in parallel. The
serial-type COMSIC, on the other hand, is accompanied by a long processing delay
according to the number of active users because the subtraction of generated
interference replicas is successively conducted in the order of the received signal
power of each user. The interference suppression effect of the parallel-type
COMSIC is inferior to that of the serial-type due to its poor channel estimation
accuracy and data detection error because channel estimation and data decision
must be performed under severely low SIR conditions at the conventional MFbased Rake receiver in the first stage. In order to improve the interference
suppression effect to the same level as that of the serial-type COMSIC, the channel
estimation accuracy must be improved and the data decision error must be
decreased to generate accurate interference replicas. Therefore, we proposed an
improved parallel-type COMSIC with iterative channel estimation (ICE) using both
pilot and decision-feedback data symbols in order to improve the interference
suppression effect to the same level as the serial-type COMSIC [48]. Our scheme is
based on recursive channel estimation at each canceling stage to improve the
accuracy of the generated interference replicas and the iterative channel
43
estimation is applied at each stage. The recursive channel estimation associated

with iterative channel estimation at each stage significantly improves the
interference suppression effect of the parallel-type COMSIC almost to the same
level as that of the serial-type.
Figure 21 is a block diagram of the parallel-type COMSIC. Composite signal
sample sequences received at two antennas are despread using MFs. At each
stage, there are K channel estimation and interference replica generation units
(CEIGUs), where K is the number of active users. Let I k,b,l be the MAI received via
the l-th path (1 l Lk) of the b-th antenna (1 b 2) of the k-th user
(1 k K)in the p-th stage (1 p P) and I(p)k,b,l be its estimated value. In the first
stage, the received signal of each antenna is directly embedded into the CEIGU as
well as the conventional MF-based Rake receiver. For the CEIGU input signal of
the b-th antenna of the k-th user in the p-th stage, (p
2), the estimated
interference replicas of all other users, ,
user's own channel,
and other multipaths of the
), which are estimated in the p vious
stage, are
weighted by the interference rejection weight control (IRWC) value, ap [47], and
removed from the signal received. It was confirmed by the some of the authors
[47] that by setting ap to less than one, the BER performance using COMSIC is
improved since the influence of the generation error of the interference replica
due to the channel estimation error and data decision error is mitigated. The
effect of IRWC is larger in the parallel-type COMSIC than in the serial-type. In our
COMSIC scheme, since the complex fading envelope (channel impulse response)
is recursively estimated and updated at each cancelling stage, the accuracy of
channel estimation and thereby that of the regenerated interference replica is
successively improved.
Figure 22 is a block diagram of the proposed CEIGU. In the proposed scheme,
the ICE using both pilot and decision-feedback data symbols after channel
decoding (hereafter forward error correction (FEC) decoding) [48] is applied to
each stage. Each CEIGU performs channel estimation, Rake combining, deinterleaving, Viterbi decoding, data decision, and replica generation of the MAI
for each user. Let kb,i(t), be the complex-valued channel impulse response of
the l-th path of the b-th antenna of the k-th user signal and(p,r)k,b,l(n)be its
estimated value after the r-th iteration (0 r R) in the p-th stage. In the first
stage, the initial channel estimate (estimated complex channel impulse response)
k,b,l(1,0)(n) of the n-th slot is computed by coherently averaging 2N pilot symbols
included in two successive slots. Then, a Rake combined data sequence is
obtained by coherently summing the Lk X B resolved signal components using
Figure 21. Block diagram of parallel-type coherent multistage interference canceller.
45
channel estimates k,b,l n ICE using both pilot and decision-feedback data
symbols after FEC decoding (hereafter ICE after FEC decoding), the Rake
(1,1)
combined sequence, d k (n, m), is de-interleaved and soft-decision Viterbi
(1,1)
decoded to recover the transmitted binary data sequence, b
(i) . The original
data sequence recovered is convolutionally-encoded and interleaved to generate
(1,1)
the binary decision-feedback data symbol sequence, d (n, m). Then, after the
data modulation of the MF output signal is removed by reverse-modulation usin
the decision-feedback data symbols, channel estimation using Nd decisionfeedback data symbols in addition to 2N pilot symbols is performed for each
slot. On the other hand, in ICE using both pilot and tentative-decision data
symbols before FEC decoding (hereafter ICE before FEC decoding), we use
binary tentative-decision data sequence after Rake combining (namely before
(1,1)
FEC decoding), sgn[d k (n, m)], instead of the decision-feedback data sequence
(1,R)
after FEC decoding in ICE after FEC decoding. L
n) and
(n, m) be t
channel estimate after the r-th iteration (1 r R), then the interference replica
of each path of each antenna is re-generated precisely as
By using the generated interference replicas, the MF input of the CEIGU of th

l-th path of the b-th antenna of the k-th user in the second stage is represented as
Furthermore, by using the interference replica generated in the previous

stage, the MF input of the CEIGU in the p-th stage (p 2) is anticipated as
where
After R iterations, data sequence d k (P,R+l) (n, m) is finally de-interleaved and

soft-decision Viterbi decoded to recover the refined decoded binary data
sequence,
Figure 22. Block diagram of CEIGU with ICE using pilot and decision-feedback data symbols.
47
VII.1.2. Experimental results

-3
The required average transmit Eb/N0 for achieving the average BER of 10 o
the parallel-type COMSIC using two-branch antenna diversity reception is plotted
in Figure 23 as a function of fD based on laboratory experiments using a fading
simulator. External Gaussian noise was added as background noise to the
receiver input. The number of active users was K = 3, and the SF of the desire
user and two-interfering users were SF = 64 and 8, respectively. The number o
stages was P = 3. The power delay profile used here was a two-path profile, each
path being subjected to independent Rayleigh fading with equal average power.
The target Eb/I0 ratio of the desired user to interfering user was set Eb/I0 = - 3
dB and - 6 dB. Thus, when Eb/I0 = - 6 dB, the received signal power of the
interfering user is 32-fold that of the desired user. The TPC delay was assumed t
be one slot. Figure 23 shows that the transmit Eb/N0 required using decision data
symbols after Rake combining with P = 3 is slightly decreased compared to that
using decision data symbols after channel decoding with P = 2 irrespective
of fD. The transmit Eb/N0 required using decision data symbols after Rake
combining with P = 2 is reduced by approximately 0.6 (0.8) and above 2.0 dB
compared to that of COMSIC using only pilot symbols and the MF-based Rake
receiver, respectively, when Eb/N0 = 3 (- 6) dB. Furthermore, the loss of th
required Eb/N0 from the single user case due to the channel estimation error and
Figure 23.
Average transmit Eb/N0 for the average BER 10-3 using parallel type comsic
as a function of fD (with antenna diversity reception, Eb = - 3, - 6 dB).
48
data decision error when Eb/N0 = - 3 dB was approximately 0.7 to 2.0 dB.
When the fD was small, the transmit Eb/N0 was slightly increased so that the fast
TPC should compensate for the drop in the signal level received due to fading.
Then, although the required transmit Eb/N0 was decreased as fD was increased
owing to the increasing interleaving effect associated with channel coding, it was
significantly degraded due to the degradation in the channel estimation tracking
ability against fast fading when fD was larger than approximately 150 Hz.
VII.2. AAAD
VII.2.1. CAAAD receiver/AAA-TD
We also investigated the interference suppression effect in the reverse and
forward links using the experimental CAAAD receiver/AAA-TD [40-45]. The
carrier frequencies were the same as those of the COMSIC. The spreading chip rate
was 4.096 Mcps and the bandwidth was 5 MHz. Figure 24 shows an overall
block diagram of the four-antenna AAAD transceiver. The receiver antenna
weights in the reverse link are generated in the CAAAD receiver block. In our
scheme, transmitter antenna weights are generated by modifying the receiver
antenna weights (i.e., performing calibrations we will be explaining later)
[40-43]. Our concern is frequency division duplex (FDD) in W-CDMA systems,
thus, the instantaneous phase and amplitude variations due to fading in the
reverse link have no correlation with those in the forward link. Noting that the
distance-dependent macroscopic propagation factors determining the average
signal power, i.e., path loss and shadowing, are not frequency dependent, we
designed the CAAAD receiver based on the following concept. The adaptive
antenna array forms an antenna beam that tracks only slow changes in the
directions of arrival (DOAs) and average powers of the desired and interfering
users, and the Rake combiner tracks the instantaneous variations in the channel
conditions to maximize instantaneous signal-to-interference plus background
noise power ratio (SINR).
We perform two calibrations to the receiver antenna weights in generating
transmit antenna weights in the forward link. We use the common array antennas
with the antenna separation of half the carrier wavelength in the reverse link.
Thus, when the DOA of the incoming signal is not 0 degrees, the phase difference
between antennas is different between different carrier frequencies. If the
generated receiver antenna weights are used directly as transmitter antenna
weights, the direction of the main lobe in the transmitter beam pattern is shifted
from the original DOA of the desired user and the directions of the beam nulls are
also shifted from those of interfering users. Therefore, in our approach, we shift
the direction of the main lobe in the transmitter beam pattern so that it coincides
Figure 24. Overall block diagram of CAAAD transceiver.
50
with the main lobe in the receiver beam pattern. We call this modification in the
transmitter beam pattern carrier frequency calibration.
On the other hand, since adaptive antenna array processing is done in the
baseband, the generated reverse link weights are reflected on the
phase/amplitude deviations in the transfer functions of the radio frequency (RF)
receiver circuitry associated with different antennas. In the transmitter, the RF
transmitter circuitry of the different antenna branches is adversely influenced by
the different transfer functions (see Figure 24). Therefore, the transmit weights
are generated by modifying the receiver antenna weights taking into account the
phase/amplitude variations in RF receiver/transmitter circuitry. This calibration i
called RF circuitry calibration. Let w(i)ideal, and w(i)R andx(i)RXbe the complex-value
antenna weight for the ideal case (no phase/amplitude errors exist in the RF
receive circuitry transfer functions), the complex-valued receiver antenna
weight generated in CAAAD, and complex-valued transfer function of the RF
receive circuitry of the i-th antenna branch, respectively. For the ideal case (i.e.,
x(i)RX=const.for all antenna branches), received signal r(i)RX should be weighted by
w(i)ideai to be combined for beam forming. However, in a real receiver, r(i)RX goes
through the RF receiver circuitry and then, weighted by w(i)R, produces w(i)R x(i)R
r(i)RX, before combining. The resultant signal must be the same as that of th
ideal case. Therefore, we obtain
The weights provided by Equation (23) cannot be directly used to form the
transmit beam because the signal weighted at the baseband stage suffers
phase/amplitude shift due to the RF transmit circuitry before transmission from
an antenna. Let x(i)TX be the transfer function of the RF transmit circuitry for the
i-th antenna branch. The signal to be transmitted from the i-th antenna is first
weighted using transmit antenna weight w(i)T and then suffers phase/amplitude
shift (equivalent by multiplication of x(i)TX). The equivalent antenna weigh
becomes, therefore, w(i)T . x(i)TX, and this must be equal to w(i)idea/. As a consequenc
we obtain
Using Equation (24), the transmit antenna weights (at the base band beam
forming stage) can be obtained from adaptively generated receive antenna
weights. Bothx(i)RXandx(i)TXcan be measured even during the operation mode. By
performing the above two calibrations on the generated receiver antenna weights
in the CAAAD receiver, the maximum gain is obtained toward the desired signal
direction and the nulls are nearly directed towards the interfering sources.
51
Figure 25 shows a block diagram of the digital beam former and Rake
combiner in the CAAAD receiver. This block comprises MFS, a beam former, PSAcoherent Rake combiner, and weight controller. The MF output signal samples of
all antennas for each resolved path are weighted by the receiver antenna weight,
which is common to all paths for Rake combining, and then combined. Since it
was reported that the angle spread among paths is within ten degrees in an urban
area with a high elevation antenna configuration such as in a cellular system [49,
50], we used the common antenna weights, i.e., common receiver beam pattern,
for all paths. The resulting composite channel gain at the beam former output is
estimated by using the pilot symbols multiplexed into the Q-channel of three
successive slots for coherent Rake combining of different resolved paths. The
receiver antenna weights in the CAAAD receiver are updated so that the mean
squared error (MSE) between Rake-combined signal and the reference signal is
minimized. As a reference signal, we used the decision-feedback data symbols
after FEC decoding in addition to pilot symbols for generating refined MSE an
improving channel estimation accuracy.
VII.2.2. Experiments
First, we investigated the effect of receiver antenna weight updating using
decision-feedback data symbols after FEC decoding in addition to pilot symbols
and performance difference for channel coding. The average BER performance
measured with the CAAAD receiver with fast TPC using convolutional coding and
turbo coding is shown in Figure 26 as a function of the average transmit E b /N 0
in the reverse link. The DOAS of the desired and interfering users were set to 0
and + 40 degrees, respectively, in a 2-user environment. The ratio of the target
Eb/I0 for fast TPC of the desired user to interfering user was E b /I 0 = - 12 dB
The performance with receiver weight updating using decision-feedback data
symbols as well as pilot symbols, and pilot symbols only are plotted as solid and
dotted lines, respectively. It was assumed that fD is 5 Hz and TCHL is 20 msec.
Figure 26 shows that the transmit Eb/N0 required for satisfying the average BER
of 10-3 and 10-6 with the CAAAD receiver using decision feedback antenna
weight updating and channel estimation is decreased by approximately 0.8 dB
compared with the case using only pilot symbols both for convolutional coding
and turbo coding. These results also confirm that for TCHL= 20 msec, the required
transmit Eb/N0 at the average BER of 10-6 of the CAAAD receiver using decision
feedback antenna weight-updating and channel estimation with turbo coding is
decreased by approximately 0.5 dB compared to the case with convolutional
coding. Thus, the results given here agree well with those from the MF-based
Rake receiver in Figure 13.
Figure 25. Block diagram of CAAAD processing.
53
Figure 26.
Average BER performance as a function of average transmit Eb/N0
(with fast TPC, 2 users).
Moreover, we evaluated, based on laboratory experiments, the forward link

performance using four-antenna AAA-TD with SIR based fast TPC. The average BE
performance of a MS using AAA-TD when fast TPC was used in the forward link
plotted in Figure 27 as a function of the average transmit E b /N 0 . The
experimental configuration of AAA-TD is described in [43]. The number of users
was four (3-interfering users) and the DOAS of the desired and interfering users
were set to D = - 50 degrees and U = -20, + 15, + 45 degrees. The BE
performance for AAA-TD (when carrier frequency calibration was applied) is
shown and the performance with one antenna transmitter is also plotted for
comparison. The ratio of the target Eb/I0 of the desired user to the interfering
users, E b /I 0 , is - 3, - 5 and - 12 dB and the single-user case is also shown.
Figure 27 clearly shows that when using the 1-antenna transmitter, as the
interfering power is increased, the BER performance is significantly degraded
i.e., error floor is observed, due to the severe MPI from the interfering users
Meanwhile, by applying AAA-TD, almost identical BER performance was
achieved even for the case of Eb/I0 = - 12 dB owing to the significant
interference suppression effect. This result demonstrates that the RF circuitry
calibration and carrier frequency calibration are working satisfactorily and that
consequently AAA-TD is very effective in decreasing strong MPI especially fro
high-rate channels.
54
Figure 27. Average BER performance of AAA-TD as a function of average transmit E b /N 0

(with fast TPC).
Next, the results of field experimental conducted near Tokyo are evaluated. In
the CAAAD receiver experiments, MS 1 (desired user) moved along the
measurement course, which was located approximately 600 to 850 m away from
the BS. The height of the BS antenna was 50 m from the ground and 120-degree
sectored antennas were used. When MS 1 moved along the measurement course,
the DOA toward the BS changed from -10 to +10 degrees. In the measurement
course, one-to-two path fading appeared in the first half of the course followed
by two-to-three path fading with an average power difference of approximately
3 dB. On the other hand, MS 2 (interfering user) was located at a fixed point 60
m away from the BS and almost in the LOS path (thus, single-path). The DOA o
MS 2 was approximately + 40 degrees. The measured BER performance is plotted
in Figure 28 as a function of the average received signal power with the average
received SIR of MS 1 as a parameter. For comparison, the results of space
diversity using MRC are also plotted (antenna separation was 10,). Figure 2
shows that the required average received signal power for obtaining the average
-3
BER of 10 is decreased by approximately 8 to 10 dB using the CAAAD receiver
compared to using space diversity.
The average BER performance measured at a MS using AAA-TD in the forwar
link is plotted in Figure 29 as a function of the average received signal power
55
Figure 28. Average BER performance as a function of average received signal power.
Field experiment.
Figure 29. BER performance as a function of average received signal power.

Field experiment. Reverse link SIR = 0 dB.
56
when the average received SIR of MS 1 is 0 dB. The transmitted SIRS of MS 1

before weighting are - 5, - 10, and - 12 dB. The BER performance of a oneantenna transmitter is also shown in the figure. The figure clearly shows that
although the BER performance with the one-antenna transmitter is severely
degraded as the transmitted SIR is decreased, the performance is significantly
improved by using AAA-TD due to beam and null steering. When AAA-TD is used,
the increase in the required transmit power at the average BER of 10-3 from th
case without an interfering user is within 5 dB when the transmitted SIR is -12
dB. These results from the reverse and forward links verify the effects of
reducing the interference from a high-rate user since the receiver and transmitter
antenna weights can track precisely the changes in the DOAS of the desired signal.
VIII. Conclusion
This paper reviewed the channel structure and spreading code assignment in
the physical layer, transport channel multiplexing into a physical channel
along with sophisticated rate matching, and the key technologies of W-CDMA
wireless access. Experimental results of the three-step cell search method
associated with TSTD exhibited fast cell search time performance, while
flexible continuous system deployment from outdoors to indoors was possible.
Pilot symbol based coherent Rake receiver exhibited efficient, i.e., lower
required Eb/I0, transmission together with fast TPC and space diversity
reception based on experiments in actual multipath fading channels. We also
verified the effectiveness of site diversity and transmit diversity in decreasing
the transmit power, thereby increasing link capacity. These experimental
results associated with the successful system level experiments showed the
superiority of W-CDMA wireless access technologies for commercial services.
Link capacity enhancing techniques such as those employing a COMSIC and
CAAAD receiver/AAA-TD and examples of BS transceiver using these techniques
were also presented with experimental results. The laboratory and field
experimental results clarified the potential of the COMSIC and CAAAD receiver to
reduce the mobile transmit power in the reverse link and that of AAA-TD to
decrease MPI with large transmit power in the forward link. These techniques ar
essential for achieving high-speed packet transmission based on the W-CDMA air
interface in an actual severe multipath fading channel.
Acknowledgements
The authors thank K. Okawa, S. Fukumoto, A. Morimoto, H. Taoka, and A.
Harada of the Wireless Access Laboratory for their contributions to the W-CDMA
experiments.
57
REFERENCES
[1] ADACHI (F.), SAWAHASHI (M.), and SUDA (H.), "Wideband DS-CDMA for Next
Generation Mobile Communication System," IEEE Commun. Mag., Vol. 36,
pp. 56-69, (Sept. 1998).
[2] DAHLMAN (E.), GUDMUNDSON (B.), NILSSON (M.), SKOLD (J.), "UMTS/IMT-2000 Based on
Wideband CDMA," IEEE Commun. Mag., Vol. 36, pp. 70-80, (Sept. 1998).
[3] SAWAHASHI (M.), HIGUCHI (K.), TANAKA (S.), and ADACHI (F.), "Enhanced Wireless
Access Technologies and Experiments for W-CDMA Communications,"
IEEE Personal Communications, vol. 7, n 6, pp. 6-17, (Dec. 2000).
[4] 3GPP RAN, 3G TS 25.301 V3.5.0, (June 2000).
[5] 3GPP RAN, 3G TS
[6] 3GPP RAN, 3G TS
[7] OKAWA (K.) and
Multi-spreading
25.211 V3.4.0, (Sept. 2000).

25.213 V3.3.0, (Sept. 2000).
ADACHI (F.), "Orthogonal Forward Link Using Orthogonal
Factor Codes for Coherent DS-CDMA Mobile Radio," IEICE Trans.
Commun., Vol. E81-B, pp. 777-784, (April 1998).

[8] 3GPP RAN, 3G TS 25.302 V3.5.0, (June 2000).
[9] 3GPP RAN, 3G TS 25.212 V3.4.0, (Sept. 2000).
[10] BERROU (C.), GLAVIEUX (A.), and THITIMAJSHIMA (P.), "Near Shannon Limit Errorcorrecting Coding and Decoding: Turbo-codes," Proc. IEEE Icc'93, pp. 1064-1070,
Geneva, (May 1993).
[11] HIGUCHI (K.), SAWAHASHI (M.), and ADACHI (F.), "Fast Cell Search Algorithm in
DS-CDMA Mobile Radio using Long Spreading Codes," IEICE Trans.Commun..
Vol. E81-B, pp. 1527-1534, (July 1998).
[12] HIGUCHI (K.), HANADA (Y.), SAWAHASHI (M.), and ADACHI (F.), "Experimental
Evaluation of 3-Step Cell Search Method in W-CDMA Mobile Radio," Proc. VTC
'2000, Tokyo, Japan, pp. 1527-1534, (May 2000).
[13] HIGUCHI (K.), HANADA (Y.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on
3-Step Fast Cell Search Method using TSTD in W-CDMA Mobile Ratio," IEICE
Technical Report, RCS99-155, pp. 99-105, (in Japanese), (Nov. 1999).
[14] ARIYAVISITAKUL (S.), "Signal and Interference Statistics of a CDMA System with
Feedback Power Control-part II," IEEE Trans. Commun., Vol. 42, pp. 597-605,
(Feb./March/April 1994).
[15] SEO (S.), DOHI (T.), and ADACHI (F.), "SIR-Based Transmit Power Control of Reverse
Link for Coherent DS-CDMA Mobile Radio," IEICE Trans. Commun., Vol. E81-B n 7,
pp. 1508-1516, (July 1998).
[16] LING (F.), "Coherent Detection with Reference-symbol Based Estimation for Direct
Sequence CDMA Uplink Communications," Proc. VTC' 93, New Jersey, pp. 400-403,
(May 1993).
[17] ANDOH (H.), SAWAHASHI (M.), and ADACHI (F.), "Channel Estimation Filter Using
Time-multiplexed Pilot Channel for Coherent Rake Combining in DS-CDMA Mobile
Radio," IEICE Trans. Commun., Vol. E81-B n 7, pp. 1517-1526, (July 1998).
58
[18] ABETA (S.), SAWAHASHI (M.), and ADACHI (F.), "Adaptive Channel Estimation for
Coherent DSCDMA Mobile Radio Using Time-Multiplexed Pilot and Parallel Pilot
Structure" IEICE Trans. Commun., Vol. E82-B n 9, pp. 1505-1513, (Sept. 1999).
[19] HIGUCHI (K.), ANDOH (H.), OKAWA (K.), SAWAHASHI (M.), and ADACHI (F.),
"Experimental Evaluation of Combined Effect of Coherent Rake Combining and
SIR-based Fast Transmit Power Control for Reverse Link of DS-CDMA Mobile Radio
IEEE J.Select. Areas Commun., Vo.18, pp.1526-1535, (Aug. 2000).
[20] HIGUCHI (K.), IKEDA (T.), FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.),
"Experimental Evaluations of High Rate Data Transmission Using
Turbo/Convolutional Coding in W-CDMA Mobile Radio" Proc. Wireless 2000, Calgary,
pp. 687-693, (July 2000).
[21] SHIBUTANI (A.), SUDA (H.), and ADACHI (F.), "Multi-stage Interleaver for Turbo Codes
in DS-CDMA Mobile Radio," Proc. IEEE APCClICC'98, pp. 391-395, (Nov. 1998).
[22] SHIBUTANI (A.), SUDA (H.), and YAMAO (Y.), "Performance of
W-CDMA Mobile Radio with Turbo Codes Using Prime Interleaver," Proc. VTC '2000,
Tokyo, Japan, pp. 1570-1574, (May 2000).
[23] GILHOUSEN (K.S), JACOBS (I.M), PADOVANI (R.), VITERBI (A.J), WEAVER (L.A), and
WHEATLEY III (C.E.), "On the Capacity of a Cellular CDMA System," IEEE Trans. Veh.
Technol., Vol. VT-40, n 5, pp. 303-312, (May 1991).
[24 WONG (D.) and LIM (T.J), "Soft Handoffs in CDMA Mobile Systems," IEEE Personal
Commun. Mag., Vol. 36, pp. 6-17, (Dec. 1997).
[25] TIA/EIA/IS-95, "Mobile Station-Base Station Compatibility Standard for Dual-Mode
Wideband Spread Spectrum Cellular System," Telecommunication Industry
Association, (July 1993).
[26] FUKUMOTO (S.), HIGUCHI (K.), MORIMOTO (A.), SAWAHASHI (M.), and ADACHI (F.),
"Combined Effect of Site Diversity and Fast Transmit Power Control in W-CDMA
Mobile Radio," Proc. VTC '2000, Tokyo, Japan, pp. 1527-1534, (May 2000).
[27] MORIMOTO (A.), HIGUCHI (K.), FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.),
"Experiments on Inter-Cell Site Diversity Using Two-step Selection Combining in
W-CDMA Reverse Link," IEICE Trans. Commun., Vol. E84-B n 3, pp. 435-445,
(March 2001).
[28] NEC, "Adjustment Loop in Down Link Power Control During Handover," TSG-RAN
Working 1 Meeting, (Oct. 1999).
[29] Nortel, "Down Link Power Control During Soft Handover," 3GPP TSG-RAN WGI
Meeting, (July 1999).
[30] FURUKAWA (H.), HAMABE (K.), and USHIROKAWA (A.), "SSDT- Site Selection Diversity
Transmission Power Control for CDMA Forward Link," IEEE J.Select.Areas Commun.,
Vol. 18, pp. 1546-1554, (Aug. 2000).
[31] HOTTINEN (A.) and WICHMAN (R.), "Transmit Diversity by Antenna Selection in CDMA
Downlink," Proc. IEEE issta'98, pp. 767-770, (Sept. 1998).
[32] FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.), "Performance Comparison of
Forward Link Transmit Diversity Techniques for W-CDMA Mobile Radio,"
Proc. PlMRC'99, pp. 1139-1143, (Sept. 1998).
[33] 3GPP RAN, 3G TS 25.214 V3.4.0, (Sept. 2000).
59
[34] ALAMOUTI (S.M), "A Simple Transmit Diversity Technique for Wireless
Communications," IEEE J. Sel. Areas Commun., Vol. 16, n 8, pp. 1451-1458,
(Oct. 1998).
[35] FUKUMOTO (S.), HIGUCHI (K.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on Space
Time Block Coding Transmit Diversity (STTD) in W-CDMA Forward Link," Submitted
to IEEICE Trans. Fundamentals. Special Issue on Spread Spectrum.
[36] DUEL-HALLEN (A.), HOLTZMAN (J.), and ZVONAR (Z.), "Multiuser Detection for CDMA
Systems," IEEE Personal Communications, pp. 46-58, (April 1995).
[37] MOSHAVI (S.), "Multi-user Detection for DS-CDMA Communications," IEEE Commun.
Mag., pp. 124-136, (Oct. 1996).
[38] TSOULOS (G.V.), BEACH (M.A.), and MCGEEHAN (J.), "Wireless personal
communications for the 21st century: European technological advances in
adaptive antennas," IEEE Commun. Mag., pp. 102-109, (Sept. 1997).
[39] COMPTON (R.T.), Jr., "An Adaptive Antenna in a Spread-spectrum Communication
System," Proc. IEEE, Vol. 66, n 3, pp. 289-295, (March 1978).
[40] HARADA (A.), TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Performance of
Adaptive Antenna Array Diversity Transmitter for W-CDMA Forward Link,"
IEEE PIMRC'99, pp. 1134-1138, Osaka, (Sept. 12-15 1999).
[41] TANAKA (S.), HARADA (A.), SAWAHASHI (M.), and ADACHI (F.), "Transmit Diversity Based
on Adaptive Antenna Array for W-CDMA Forward Link," Proc. CDMA International
Conference (CIC)'99, pp. 282-286, Korea, (Sept. 1999).
[42] HARADA (A.), TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on
Adaptive Antenna Array Transmit Diversity in W-CDMA Forward Link," Proc. CDMA
International Conference (CIC)'2000, pp. 47-51, Korea, (Nov. 2000).
[43] TAOKA (H.), TANAKA (S.), IHARA (T.), and SAWAHASHI (M.), "Experiments on Adaptive
Antenna Array Transmit Diversity with Carrier Frequency Calibration in Transmit
Power-Controlled Forward Link for W-CDMA Mobile Radio," Proc. 3G Wireless and
Beyond, San Francisco, (May 2001).
[44] TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Pilot Symbol-assisted Decisiondirected Coherent Adaptive Array Diversity for DS-CDMA Mobile Radio Reverse
Link,"IEICETrans. Fundamentals, Vol. E80-A, pp. 2445-2454, (Dec. 1997).
[45] TANAKA (S.), HARADA (A.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on
Coherent Adaptive Antenna Array Diversity for Wideband DS-CDMA Mobile
Radio," IEEE J.Select.Areas Commun., Vol. 18, pp. 1495-1504, (Aug. 2000).
[46] SAWAHASHI (M.), MIKI (Y.), ANDO (H.), and HIGUCHI (K.), "Pilot Symbol-Assisted
Coherent Multistage Interference Canceller Using Recursive Channel Estimation
for DS-CDMA Mobile Radio," IEEICE Trans. Commun., Vol. E79-B, pp. 1262-1270,
(Sept. 1996).
[47] SAWAHASHI (M.), ANDOH (H.), and HIGUCHI (K.), "Interference Replica Weight
Control for Pilot Symbol-Assisted Coherent Multistage Interference Canceller
Using Recursive Channel Estimation in DS-CDMA Mobile Radio," IEICE Trans.
Fundamentals, Vol. E81-A, pp. 957-972, (May 1998).
60
[48] OKAWA (K.), HIGUCHI (K.), and SAWAHASHI (M.), "Parallel-type Coherent Multi-stage
Interference Canceller with Iterative Channel Estimation Using Both Pilot and
Decision-feedback Data Symbols for W-CDMA Mobile Radio," IEICE Trans.
Commun., Vol. E84-B n 3, pp. 446-456, (March 2001).
[49] KOZONO (S.) and SAKAGAMI (S.), "Correlation coefficient on base station diversity
for land mobile communication system," IEICE Trans Commun., Vol. J70-B,
pp. 476-482, (in Japanese), (Apr. 1987).
[50] MARTIN (U.), "Spatio-temporal radio channel characteristics in urban macrocells,"
IEEE Proc.- Radar, Sonar Navig., Vol. 145, (Feb. 1998).
[51] IMT-2000 Study Committee, AIF/SWG2-25-4, (July 1998).
Chapter 3
Multicarrier CDMA techniques for

future wideband
wireless networks
M. Hlard and R. Le Gouable
France Telecom R&D
J.-F. Hlard and J.-Y. Baudais

INSA/LCST, France
I. Introduction
Future multimedia services (Internet, video transmission, data transfer, ...) will
require the transmission of very high data rates over broadband radio channels.
In order to meet the quality of service (QOS) requirements of these future
multimedia applications, the multiple access techniques of next generation
cellular systems should offer flexibility and especially very high spectral
efficiency.
Spread spectrum has been very successfully used by the military services for
decades and more recently has taken on a significant role in cellular and personal
communications [1, 2]. Advantages of spread spectrum techniques are widely
known: immunity against multipath distortion and jamming, low transmitted
power, no need for frequency planning and last but not least, Code-Division
Multiple Access (CDMA) capabilities [3]. As a result of the discussion in the last
decade, CDMA will play an important role in the next generation of cellular
systems, i.e. in UMTS and in IMT-2000 [4]. Thus, Wideband CDMA is a leading
candidate for third generation wireless access. Based on direct sequence spread
spectrum with a chip rate of 3.84 Mchip/sec, the transmitted signal occupies a
bandwidth of about 5 MHz. It supports circuit and packet data access at nominal
peak data rate equal to 384 Kbit/sec for macro cellular environments and to
2 Mbit/sec for indoor environments. However, we already know that wireless
Internet usage is likely to be downlink limited. Indeed, for data services, peak bit
62
Multicarrier CDMA techniques
rate is very important in determining overall system performance because of the

highly bursty nature of Internet traffic.
On the other hand, the multi-carrier technique, well known under the
acronym OFDM, which stands for Orthogonal Frequency Division
Multiplexing, has been receiving widespread interest for wireless broadband
multimedia applications over the last decade. The main advantages of this
technique are its robustness in the case of frequency selective fading channels,
its capability of portable and mobile reception and its flexibility. With this
technique, the serial high rate baseband stream is distributed over many
closely spaced orthogonal subcarriers. Due to the large number of subcarriers,
the symbol duration is much higher than the channel time dispersion, which
minimizes the Inter-Symbol Interference (ISI). Furthermore, residual ISI is
suppressed thanks to the insertion of a guard interval larger than the channel
time delay at the beginning of each symbol.
The OFDM technique was initially selected in 1990 in the European Digital
Audio Broadcasting (DAB) standard [5, 6]. Its further prominent successes were
in 1995 and 1998 as it was selected for the European Digital Video Broadcasting
(DVB-T) [7] and in three broadband wireless indoor standards, namely ETSIHiPERLAN/2 [8], American IEEE-802.11a [9] and Japanese MMAC. Furthermore,
OFDM is becoming the basis of a world standard for asymmetric digital
subscriber lines under the acronym of DMT for Digital Multitone [10].
The advantages and success of multi-carrier modulation and spread spectrum
technique motivated many researchers to investigate the suitability of the
combination of both techniques. This combination known as Multi-CarrierSpread-Spectrum (MOSS) benefits from the main advantages of both schemes:
high spectral efficiency, high flexibility, multiple access capabilities, narrowband interference rejection, simple one-tap equalization, etc., ... Three different
concepts based on the combination of multi-carrier modulation with DS-CDMA
have been introduced in 1993. They are known as "Multi-Carrier CDMA"
(MC-CDMA) also called OFDM/CDMA [11-14], Multi-Carrier Direct Sequence
CDMA (MC-DS-CDMA) [15] and Multi-Tone CDMA (MT-CDMA) [16]. The main
differences between them are in the spreading that is carried out either in time or
in frequency domain, in the frequency mapping and in the detection techniques.
Because of the limited frequency bandwidth on the one hand and the
limited power of mobile units on the other hand, spectral and power efficiency
of future systems should be as high as possible. This explains why the topic
of MC-SS has recently received widespread interest from researchers, judging
from the number of papers at international conferences. During the last years,
deep system analysis and comparison of MC-SS and mainly MC-CDMA with
DS-CDMA have been performed demonstrating the superiority of MC-CDMA
63
[17-19].The main topics studied are coding and modulation aspects,

multi-user detection techniques as interference cancellation strategies,
synchronization and channel estimation aspects, realization and
implementation issues.
This paper is mainly devoted to the study of MC-CDMA systems for the
downlink of high rate cellular networks over UMTS channels. The performance of
different mono-user and multi-user detection techniques are compared. It is
shown that a new linear detection technique named Global Minimum Mean
Square Algorithm (GMMSE) offers very good performance mainly for non-full
load systems. Furthermore, the application of channel coding as convolutional
and turbo codes in synchronous MC-CDMA systems, in conjunction with this
GMMSE detection technique is examined.
The paper is organized as follows: in section II, the three types of MC-SS
schemes are reviewed. Their advantages and disadvantages in terms of
transmitter and receiver structures, spectral efficiency and power efficiency
are discussed. Then, in the following sections, the performance of MC-CDMA
systems is studied in the case of a downlink transmission. Section III deals
with analytical expressions to evaluate the performance of different detection
techniques such as Maximum Ratio Combining (MRC), Equal Gain Combining
(EGC) and Minimum Mean Square Error (MMSE). Monte Carlo simulations
confirm the analytical results on a Rayleigh channel. After a brief description
of the interference cancellation techniques in section IV, a new Global
Minimum Mean Square Error (GMMSE) detection technique recently published
in [20] is presented in section V and compared to the Parallel Interference
Cancellation detection scheme. The UMTS channel modelling and system
parameters used for simulations are described in section VI. Different results
are given for a synchronous MC-CDMA system with various detection
techniques, in order to look for the system that will offer the best trade-off
between performance and complexity. With regard to the channel coding,
convolutional codes and turbo-codes are considered. Finally, section VII
summarizes the results and draws together the conclusions.
II. Multi-carrier spread spectrum concepts

II.1.
MC-CDMA
The first concept, known as OFDM/CDMA or more often as MC-CDMA, is based on

a serial concatenation of Direct Sequence spreading with Multi-Carrier
Modulation. The MC-CDMA transmitter spreads the original data stream over
different subcarriers in the frequency domain using a given spreading code. The
64
Figure 1.
MC-CDMA transmitter scheme and power spectrum of the transmitted signal.
effect of spreading is that different users can have access to the same carriers in
a CDMA manner. The separation of the user's signals is then performed in the
code domain. The advantage of MC-CDMA in comparison with DS-CDMA is that the
spreading can be adapted to the frequency selective behaviour of the channel.
Simple methods for signal detection in the frequency domain as one-tap
equalizer per carrier can be used.
Figure 1 shows the MC-CDMA transmitter of the jth user and the power
spectrum of the transmitted signal.
The data symbol xj. (t) of the user j is transmitted in parallel over Nc
subcarriers, each multiplied by one chip ck, j of the spreading code Cj (t) = [c1,j.,
c 2,j , ..., cLmc, j] assigned to user j. In this figure, the length Lmc of the spreading
code is equal to the number NC of subcarriers but this is not mandatory. As a
consequence, the MC-CDMA systems offer an additional degree of freedom, and
actually the number Nc of subcarriers is chosen to guarantee frequency nonselective fading over each subcarrier.
65
The expression of the transmitted signal Sj (t) of user j during the time
interval [0, TX] is:
where Tx is the data symbol duration which is in this case equal to the OFDM
symbol duration, xj the data symbol transmitted during the signalling interval
[0, Tx], P (t) the pulse shaping waveform which is generally rectangular and f0
is the carrier frequency.
Practically, the MultiCarrier modulation and demodulation is easily carried
out in the digital domain by performing IFFT and FFT operations. Furthermore the
insertion between adjacent MultiCarrier (MC) symbols of a guard interval A,
longer than the delay spread of the impulse response of the channel, guarantees
the absence of Inter Symbol Interference (ISI). In this case, the resulting
spectrum of the transmitted signal satisfies the orthogonality condition with the
minimum frequency separation equal to 1/TX = 1/(T' X -), where T'x is the new
MC symbol duration and Tx is the useful duration of this symbol. In the receiver,
after direct FFT and possibly de-interleaving, the received sequence is
"equalized" in the frequency domain. Therefore, the MC-CDMA receiver can
always employ all the received signal energy spread in the frequency domain.
Undoubtedly, this is the main advantage of the MC-CDMA scheme compared to a
DS-CDMA Rake receiver that has difficulties in making full use of the received
signal energy scattered in the time domain.
For a synchronous system as the downlink mobile radio communication
channel, the application of orthogonal codes such as Walsh-Hadamard codes
guarantees the absence of Multiple Access Interference (MAI) in a gaussian
channel. However, in non-ideal channels with frequency selective fading due to
multipath propagation, the orthogonality between the signals of the different
users is lost and MAI occurs. To combat the channel fading and thus the MAI, a
multitude of detection techniques was proposed. They can be classified as either
single-user detection (SD) or multi-user detection (MD). Single-user detection is
performed by one tap equalization to compensate for the phase and amplitude
distortions caused by the mobile radio channel. The one tap equalizer is simply
one complex-valued multiplication per subcarrier. Basic SD techniques are
Maximum Ratio Combining (MRC), Equal Gain Combining (EGC), Orthogonal
Restoring Combining (ORC) or Minimum Mean Square Error (MMSE)
equalization. This last technique, based on the MMSE criterion applied
independently to each subcarrier achieves better performance [21, 25] as we
66
will see in section III. Another way to mitigate the MAI is to optimize the
spreading sequence selection within a given spreading sequence family as
presented in [26].
With the aim to improve the performance of the receiver still further, Multiuser Detection (MD) can be processed, where the a priori knowledge about the
spreading codes of the interfering users is exploited in the detection process.
Indeed, the receiver jointly detects the other active user signals in order to
mitigate the MAI. Multi-user detection methods can be divided into Interference
Cancellation (IC) and Joint Detection (JD). The principle of IC is to detect the
information of the interfering users and to reconstruct the interfering
contribution in order to subtract it from the received signal.IC can be performed
parallel for all interfering users with Parallel Interference Cancellation (PIC)
detectors, or successively with Successive Interference Cancellation (SIC)
detectors where only the strongest interferer remaining after the previous IC
stage is cancelled.
The optimal detector applies Joint Detection (JD) with Maximum Likelihood
Sequence Estimation (MLSE) or Maximum Likelihood Symbol-by-Symbol
Estimation (MLSSE), respectively. Since the complexity of MLSE and MLSSE
receivers grows exponentially with the number of users, their use is limited in
practice to applications with a small number of users.
Most MC-CDMA systems were considered for optimum use of the available
bandwidth of the synchronous downlink of a cellular network using orthogonal
codes as Walsh-Hadamard codes. For the uplink, another scheme called Spread
Spectrum Multi-Carrier Multiple Access (SS-MC-MA), which is derived from the
MC-CDMA concept, has been proposed in [27]. In this case, each user k,
exclusively transmits on a set of L subcarriers out of a total of NC carriers.
II.2. MC-DS-CDMA
As shown in Figure 2, the MC-DS-CDMA transmitter modulates the substreams on

subcarriers with a carrier spacing proportional to the inverse of the chip rate to
guarantee the orthogonality between the spectrums of the substreams after DS
spreading [15, 28].
First, the data stream is converted into parallel low rate substreams before
applying the DS spreading on each substream in the time domain and
modulating onto each subcarrier. This scheme was mainly investigated for an
asynchronous uplink communication channel.
The transmitted signal Sj (t) of user j is given by:
67
Figure 2.
MC-DS-CDMA transmitter scheme and power spectrumof the transmitted signal.
If the spreading code length Lmd is inferior or equal to the number of subcarriers
Nc, a single data symbol is spread in the time domain. By using a high number
of subchannels, this concept benefits from time diversity. However, due to the
frequency non-selective fading per subchannel, frequency diversity can only be
exploited if channel coding with appropriate interleaving or subcarrier hopping
is used, or if the same information is transmitted in parallel on several
subcarriers [29]. In [30], a MC-DS-CDMA scheme with a subcarrier spacing larger
68
than the inverse of the chip duration is proposed in order to increase the
frequency diversity of the system.
In reception, the MC-DS-CDMA system is composed of Nc coherent receivers
for data detection.
II.3. MT-CDMA
Concept 3, known as Multi-Tone-CDMA (MT-CDMA) uses the same data mapping

and spreading as concept 2. The MT-CDMA transmitter, depicted in Figure 3,
spreads the Serial to Parallel converted data stream using a given spreading code
in the time domain [16]. Thus, the Nc parallel data symbols fulfil the
orthogonality requirements before DS spreading. However, after the DS
spreading, the resulting spectrum of each subcarrier no longer satisfies the
orthogonality condition, which generates Inter-Carrier Interference (ICI). On the
other hand, the tight subcarrier spacing enables the use of spreading codes with
a length Lmt that is longer by a factor of approximately Nc than the length of the
spreading code of the DS-CDMA scheme. Therefore, at the expense of higher ICI,
under certain conditions, the system can accommodate more users than the DSCDMA scheme.
The expression of the transmitted signal Sj (t) of user j is:
with
which is identical to the expression of the MC-DS-CDMA transmitted signal given
by equation (2) except that the subcarrier separation is equal to l/(N c TX) instead
of Lmd/(Nc Tx). The MT-CDMA receiver is composed of Nc Rake combiners, each
of which has the same structure as the DS-CDMA Rake receiver. The MT-CDMA
scheme suffers from inter-subcarrier interference, while the capability to use
longer spreading codes, compared to the spreading codes assigned to a CDMA
scheme, results in the reduction of Self Interference (SI) and MAI. In a channel
where this improvement is dominant, the MT-CDMA scheme can outperform the
DS-CDMA scheme [17]. This concept was mainly investigated for an asynchronous
up-link scenario.
II.4. Comparison of different systems

In Table I, the main features of these three schemes are given in function of Tx,
the input symbol duration.
Advances in
MT-CDMA transmitter
UMTS
technology
Figure 3.
scheme and power spectrum of the transmitted signal.
Table I . System features comparison.
I
I
Svstem
Spreading
MC-CDMA
MT-CDMA
Time domain
Time domain
Spreading length
Lm
Lmd-Lm
Lmr sNc . Lmc
Symbol duration
at subcarrier
Tx
N, . T, I L-
N, . Tx
Subcarrier
separation
1 I TI
Lmd J (N, . T,)
IlN,.T,
TI
( N , TJ J L ,
(Nc . T,) J Lmt
( N , t l)L,/
0 3 2 Lmr+N-1J
Chip duration
I MC-DS-CDMA I
I Frequency domain I
Required
bandwidth
(main lobe)
Reception
techniques
Main
applications
I
1
( N , + 1 ) T,
- MRC, EGC, ORC,

MMSE
-PIC, SIC, MLSE
-
Downlink
-Quasi- synchm
nous uplink
I
I
( N c . TI)
- Coherent
Asynchronous
-uplink
4NJJ
Roll-off factor = 0
- Rake combiner
-MlMOequalizer
-Asynchronous
plink
69
70
Since 1993, especially the first and the second schemes have been studied
deeply. The capacity of the three systems is equivalent if Lmc Lmd (Lmt/Nc)
Nc. When a Nyquist filter with a small roll-off factor is used in the MT-CDMA
scheme, the MC-CDMA, MC-DS-CDMA and MT-CDMA systems require almost the
same frequency bandwidth to transmit the same bit rate 1/Tx. Deep system
analysis and comparison with DS-CDMA have been performed showing the
superiority of MC-CDMA [17-19]. In particular, for the synchronous downlink, the
MC-CDMA system with MMSE detection outperforms all other schemes and offers
the best trade-off between the spectrum and the power efficiencies.
As pointed out in the introduction, wireless Internet access is expected to grow
rapidly in the near future. This means that spectrum efficiency of the downlink
becomes crucial for these high-speed data services. In the following sections, the
performances of MC-CDMA systems are therefore considered for the downlink of
the future wideband wireless networks.
III. Performance analysis of MC-CDMA systems with

single-user detection techniques
III.1. MC-CDMA transmitter and receiver
The block diagram of the MC-CDMA transmitter and receiver considered is
depicted in Figure 4 for the downlink. Each data symbol xnj assigned to user j,
j = 1, ..., Nu and transmitted during the symbol interval n is multiplied with its
user specific Walsh-Hadamard spreading code C j (t)=[c 1 , j , c2,j ,..., c Lmc,j ] T of
length L , where [.]T denotes matrix transposition. L corresponds to the
Figure 4. Studied MC-CDMA transmitter and receiver.
71
bandwidth expansion factor and is equal to the maximum number of

simultaneous active users. The j-th column vector of the LC x LC matrix C
corresponds to the spreading code C of the user j.
The vector of the data symbols transmitted during the n-th Orthogonal
Frequency Division Multiplexing (OFDM) symbol by all the users can be
n
n
n
n
n
T
n
written X = [x 1 , x 2, , x j, ..., x Lmc] , with x j = 0 when user j is inactive.
Since we consider the synchronous downlink of an MC-CDMA system, the
different data modulated spreading codes of the Nu users can be added before
Serial-to-Parallel (S/P) conversion. Furthermore, the Nu user signals are
supposed to be transmitted with the same power. The number Nc of subcarriers
which are QPSK modulated is chosen equal to the spreading code length Lmc. For
this study, frequency non-selective Rayleigh fading per subcarrier and time
invariance during one OFDM symbol are assumed. The absence of Intersymbol
Interference is also guaranteed by the use of a guard interval longer than the
delay spread of the impulse response of the channel. Based on these assumptions
and considering time and frequency interleaving, the complex channel fading
coefficients are independent for each subcarrier and can be estimated for the
subcarrier k by hk = k e i k. The signal received after the inverse OFDM operation
(serial to parallel conversion and direct FFT) and de-interleaving can be
expressed as:
where the Nc x Nc diagonal matrix H = diag{h 1 , ..., hNc} describes the complex
channel frequency response and N = [n1, n2, ..., nNc]T is the vector containing
the Additive White Gaussian Noise (AWGN) terms with nk representing the noise
term at the subcarrier k with variance given by
=N2=E{\nk\2},k = 1, ..., Nc.
After equalization, the received signal can be written as:
In this section, the Nc x Nc matrix G, which represents the complex

equalization coefficients is diagonal with G = diag{g 1 , ..., g N c }. The different
coefficients gk can be derived from the channel estimation which is based on
known transmitted pilot symbols inserted between the data carriers. Finally,
after despreading and threshold detection, we obtain the detected data symbol
xj., which corresponds to the sign of the scalar product of the received vector
Y and the specific spreading code Cj as:
72
where X represents the useful signal part, the MAI and the noise term.
III.2. Single user detection systems: description and performance

analysis
III.2.1. Description of different single user techniques
The calculation of the bit error probability Pb is based on the law of the large
numbers, i.e the spreading code length Lmc and then the number of subcarriers
Nc must be sufficiently high. In this case, owing to the fact that the different
multiple access interference terms can be considered independent and using the
central limit theorem, the interference term and the noise term can be
estimated by an additive zero mean Gaussian noise with a variance respectively
equal to 2 and 2. Furthermore, as the interference term and the noise term are
independent, the sum of both has a zero mean Gaussian distribution with the
variance equal to the sum of the variances. Therefore, the bit error probability
can be estimated with:
where E is the useful signal energy.

The average energy of the data bits of the different Nu user signals received
is supposed to be identical and equal to Eb. Furthermore, thanks to the
interleaving, the complex channel fading coefficients hk are considered
2
independent for each subcarrier and normalized such as E[|hk| ] = 1. Thus, the
random distributed variable equal to the product of hk with gk has the first and
2
2
second moment E[g h] and E[g h ] respectively independent of the index k. So,
when applying the law of large numbers, the useful in-phase signal energy is
given by:
73
In order to simplify the notation, we consider hereafter only the in-phase

signal. Of course, the calculation is the same for the quadrature component.
2
In the same way, the variance of the multiple access interference term for the
in-phase signal is:
Since the codes are orthogonal, half of the products ck, ick, j, i j, are equal to
1/Lmc
and the other half are equal to -\/Lmc
. Then, the variance of the multiple
mc
mc
access interference term can be written as:
and the variance of the additive zero mean Gaussian noise is given by:
In the following, the values of E, 2 and 2 are presented for various single
user detection techniques when the fading of the channel is Rayleigh distributed.
Maximum Ratio Combining technique (MRC)
In the single user case, MRC is the optimum diversity combining technique. The
corresponding equalization coefficients are:
where * stands for complex conjugation.

However, in a multi-user scenario, the multiplication by the conjugate
complex channel coefficients results in enhanced MAI. The different quantities
are equal to:
74
Equal Gain Combining technique (EGC)

With EGC, only the phase shift is corrected. So, in fading channels the
orthogonality of the Walsh-Hadamard spreading codes gets lost, resulting in MAI.
The equalization coefficients are:
The variances of the multiple access interference term and the additive zero
mean Gaussian noise are respectively:
Orthogonality Restoring Combining technique (ORC)

This technique, also called Zero Forcing (ZF), inverses the channel transfer
function and thus restores the orthogonality between the users.
As the MAI is completely eliminated, performance does not depend on the

number of active users. The drawback of ORC is that for small amplitude of hk,
the noise level is enhanced. For this detection technique, the analytic estimation
of the bit error probability is not possible with this approach because the
calculation of the variance 2n does not converge.
Minimum Mean Square Error technique (MMSE)
Among all these single-user detection techniques, MMSE equalization offers the
best results. It minimizes the mean square value of the error k between the
signal sk transmitted on subcarrier k and the assigned output yk of the equalizer.
The equalization coefficients based on this MMSE criterion applied independently
per carrier are equal to:
where c is the subcarrier signal to noise ratio.
75
To calculate the bit error probability, we set z = {gkhk} and x = \hk\ 2.

As E[\h k \2] = 1, the probability density of x is p(x) = e -x . Therefore, we
obtain:
with
As already reported in Section III. 1, the matrix G is diagonal for all these
basic single user detection techniques which means that the sequence received
is equalized by using a bank of Nc adaptive one tap equalizers which results in a
low complexity equalizer.
III.2.2. Analytical and simulation results with single user detection
techniques
The analytical and simulation results with MRC, EGC and MMSE detections are
presented in Figure 5 for a Rayleigh channel. The number Nu of active users is
equal to the spreading code length Lmc (full user capacity) which is also equal to
the number Nc= 64 of subcarriers.
The Matched Filter (MF) bound or "limit bound" is given as reference. It
corresponds to the BER obtained in the case of MRC detection for a MC-CDMA
system with only one user and consequently not affected by MAI.
It can be seen that the analytical results are really close to the simulation
results, particularly for MRC and EGC detections. For MMSE detection, there is a
76
Figure 5. Single user detection techniques: analytical

and simulation results; full load system: Nu = Lmc = Nc = 64.
slight difference between analytic and simulated results because the

contribution of the multiple access interference is not exactly Gaussian.
Indeed, for the analytic approach, the number of interfering terms is not large
enough to obtain a Gaussian resulting signal according to the central limit
theorem. Nevertheless, the MMSE outperforms the other techniques avoiding an
excessive noise amplification for low signal to noise ratios while restoring the
orthogonality among users for large signal to noise ratios. The potential of
MMSE already pointed out in many references [21, 25] is there confirmed.
However, this MMSE equalization per carrier method is not optimal, since it
does not take into account the despreading process and thus does not minimize
the mean square error at the input of the threshold decoder. In order to obtain
better performance, a new method based on a global implementation of the
MMSE criterion is presented in Section V.
IV. Multi-user detection techniques

There are two major types of Multi-User (MU) detectors MU detectors based
either on the Maximum Likelihood algorithm or on the Interference
Cancellation (IC). They both need the knowledge of all the spreading codes used
and consequently are more complex to implement than SU detectors. However,
a significant gain is expected with the use of the MU detectors. The MLD
(Maximum Likelihood Detector) (MLD) may be carried out by a Viterbi
decoder. Nevertheless, the MLD detector has quite a high complexity that
exponentially increases with the number of active users [31]. So, in order to
handle a large number of users, receivers can implement suboptimal nonlinear
interference cancellation (IC) techniques with lower complexity. We will
77
describe the two types of IC detectors based on either a serial (SIC) or a parallel
(PIC) interference cancellation [32, 33].
The aim of each of these IC detectors is to estimate the contribution of
each user in the MAI term and to suppress it. The process may be performed
iteratively in multiple detection stages. The useful received signal is first
equalized by a SU technique and then despread by each code. Since the
downlink is considered synchronous, an Inverse Fast Hadamard Transform
(IFHT) may be carried out to reduce the receiver complexity.
IV.1. Successive Interference Cancellation
The SIC detector first detects the most powerful interfering user and then cancels
its contribution from the received signal. The second strongest interferer is then
cancelled and so on. The processing may be repeated for a few or for all users.
A complete detector would consider all users, but only the interferers stronger
than the useful one are commonly suppressed, SIC detector is generally used
when the power of some users are higher than the power of the useful user. Since
processing one supplementary stage leads to an additive time delay, a trade-off
between the number of stages and the total acceptable delay has to be found.
Figure 6 shows the scheme of one stage SIC detector where the strongest
interfering user is detected among the estimations x1 ... XNU and then cancelled
from the received signal. The process is carried out iteratively until the remained
interferers are considered insignificant. The resulting signal is finally despread.
The data detection may be hard or soft.
Figure 6. One stage SIC detector schemeeme..
IV.2. Parallel Interference Cancellation

The Parallel Interference Cancellation (PIC) structure is based on an estimation
of the total interference due to the simultaneous other users in order to remove
it from the received signal. Figure 7 shows the scheme of a PIC detector, where
the contribution of all interfering users is cancelled in parallel, reducing the
78
me.Figure 7. One stage PIC detector scheme..
time delay of a SIC detector. The expression of this iterative system for the
m-th stage and the j-th user is given by the following:
with the expression of the initial stage given by:
The received signal is first equalized by a SU technique, then it is despread

by each code. An Inverse Fast Hadamard Transform (IFHT) can be
implemented since the system is synchronous. As for the SIC detector, data
detection may be either hard or soft. After detection, the data is spread again,
tapped by the estimated channel coefficients H and then subtracted from the
signal received. Finally, the resulting signal with lower MAI term is then
equalized, despread and detected. We can note that the second equalizer
structure (G[m]) may be different from the first (G [m - 1]).
79
V. A new global minimum mean square error

(GMMSE) detection technique
The aim of this new method, which has been patented [34] and named the Global
MMSE algorithm, is to minimize the mean square error between the transmitted
symbol x. and the estimated symbol xj. [20]. Let
T
NC
W j= [w j, wj , ..., Wj ] be the optimal weighting vector. The estimated symbol

of the j-th user is:
According to Wiener filtering, the optimal weighting vector is:
where R, R is the autocorrelation matrix of the received vector R and R,xj is the
cross-correlation vector between the desired symbol x. and the received signal
vector R. These terms are equal to:
The optimal weighting vector can be written as:
The subcarrier noises have the same variance and are independent. Thus,
E{NN*T} = 2N,.1, where / is the identity matrix. Since the user signals have the
same power (E{x j 2 } = ES) and are independent, we can write E{XX*T} = Es.A,
where A = {aij} is a diagonal matrix with the term ajj = 1 if the user j is active
and ajj = 0 if the user j is inactive. Then, the equalization coefficient matrix is:
In the full load case (Nu = Lmc) and only in that case, the quantity C.A.CT is
equal to the identity matrix and the equalization coefficients matrix G is a
diagonal matrix with the k-th subcarrier equalization coefficient equal to the
former Equation (19).
80
On the other hand, when the capacity is not full (Nu < Lmc), the equalization
coefficient matrix G is no more diagonal. In that case, as will see in the last
section, the Global MMSE (GMMSE) algorithm outperforms the MMSE per carrier
algorithm, since it minimizes the decision error taking into account the
de-spreading process instead of minimizing the error independently on each
subcarrier.
VI. Comparison of different detection schemes:

simulation results
VI.1. System and simulation parameters
VI.1.1. System parameters
Results presented in this final part concern simulation results over channels that
are representative of vehicular test environments in UMTS standard. For our study,
we considered BU (Bad Urban) and HT (Hilly Terrain) channels as defined in GSM
standard [35] and two UMTS channels defined for vehicular environments [36].
In our simulations, the propagation channels consist of 6 or 12 power loss paths
and lognormal shadowing of zero mean with classical Jakes Doppler spectrum on
each path. Table II describes the tapped-delay line parameters for BU, HT, VA and
VB channels as defined by [35] and [36].
Table III gives the carrier frequency F0, the maximum velocity Vm of the
mobile user, the resulting maximum Doppler frequency fDmax and the delay
spread Tm for each channel.
Table II. Tapped delay line parameters of BU, HT, VA and VB channels.nnels..
BU: Bad Urban
HT: Hilly Terrain
Vehicular A
Vehicular B
Delay sec Power dB Delay sec Power dB Delay sec Power dB Delay sec Power dB
0.0
-7
0.0
-10
0.0
0.0
0.2
-3
0.2
-8
0.31
-1
0.3
0.0
0.4
-1
0.4
-6
0.71
-9
8.9
-12.8
0.8
0.8
-4
1.09
-10
12.9
-10.0
1.6
-2
1.6
1.73
-15
17.1
-25.2
2.2
-6
2.0
2.51
-20
20.0
-16.0
3.2
-7
2.4
-4
5.0
-1
15.0
-8
6.0
-2
15.2
-9
7.2
-7
15.8
-10
8.2
-10
17.2
-12
10.0
-15
20.0
-14
-2.5
81
Table HI. Channel and system parameters.
HT
VA
VB
Channel
BU
FO
1.8 GHz
1.8 GHz
2 GHz
2 GHz
1 m/sec
27.7 m/sec
27.7 m/sec
27.7 m/sec
6 Hz
166.7 Hz
185 Hz
185 Hz
Tm
10. sec
20. sec
2.5 sec
20. sec
20. sec
20. sec
5. sec
20. sec
Tx
128. sec
128. sec
33.33 sec
133.33 sec
256
256
256
1024
11
0.793 bit/sec/Hz
0.793 bit/sec/Hz
0.797 bit/sec/Hz
0.797 bit/sec/Hz
2 MHz
2 MHz
7.68 MHz
7.68 MHz
Dmax
Nc
The last parameters of the table depending on the system are given for each
channel:
-
the guard interval A that is equal or longer than Tm,

the useful OFDM symbol duration Tx,
the number Nc of subcarriers,
the total spectral efficiency , given for a system including a
1/2 rate channel code,
- the bandwidth W of the transmitted signal.
The characteristics of the simulated MC-CDMA system are the following:
- the length Lmc of the Walsh Hadamard spreading code is equal to 64,
- different data per user per OFDM symbol for BU, HT and VA channels and
16 for VB channel,
- the maximum number of users Nu is equal to 64 when the system is
full load,
- the subcarriers are QPSK modulated,
- a random frequency interleaving is carried out over one OFDM symbol,
- all users have the same power.
Channel estimation and synchronization are assumed to be perfect and based on
the insertion of pilot symbols between the data. The arrangement of these pilots
must guarantee an optimum sampling of the channel transfer function in time
and in frequency, which depends on the coherence bandwidth and the coherence
time of the channel. Nevertheless, in this case, we estimate that a good sampling
of the channel response can be reached with the insertion of one pilot every Np
= 12 carriers, which induces a loss of spectral efficiency equal to 8.3%. So, as
we use a channel code rate R = 1/2 associated to QPSK modulation (M = 2 bits
per symbol), the total spectral efficiency is equal for a full load system to:
82
Furthermore, all users are considered to have the same power, which means
that power control may be implemented. For the comparison of the different
detection systems, Bit Error Rate (BER) performance is provided as a function of
Eb/N0 where Eb is the energy per useful bit. All results have been obtained
through simulation with the Monte Carlo method. For most results,
2 limit curves are provided:
a lower limit called "limit limit bound" that stands for a system with only
one user and a matched filter (MRC detection) and consequently without
MAI,
a upper limit that corresponds to a "full load" system with the maximum
MAI.
In order to evaluate the performance of the MC-CDMA system on the

downlink, we first study the degradations due to MAI. For all systems including
interference cancellation techniques, a one stage PIC detector with hard decision
is implemented. Therefore, we first obtain a set of curves versus Nu the number
of users. Then we introduce channel coding and analyse the behaviour of a
complete system.
VI.2. Comparison between different detectors without channel coding
We will compare performance of 3 different detectors: one Single User (SU)
detector, one Multi User (MU) detector and a GMMSE detector.
For performance of a system including a SU detector, only the results of
the MMSE algorithm (Minimum Mean Square Error algorithm) are presented.
Indeed, as we have already seen in Section III.2.1, the SU MMSE detector
revealed to be the best single user detector [22]. For the MU detector, we
only present results with one stage of cancellation PIC detector where the IC
process is carried out with the MMSE technique. In fact, PIC and SIC detectors
provide equivalent results for quite similar complexity. Carrying out a
second stage of PIC does not largely improve performance; indeed, the gain
remains under 1 dB for a higher complexity [23]. In order to evaluate the
influence of MAI, we first present results without channel coding for most
channels.
VI.2.1. Full load system performance
Figure 8 illustrates the BER performance of SU and MU detectors for a full load
system corresponding to 64 users. The curves obviously demonstrate how
83
Figure 8. Performance of the SU MMSE and MU PIC detectors

for a full load system over BU, HT and VB channels.
efficiently MAI is fought by a MU detection technique that widely outperforms the

SU MMSE detector. Nevertheless, at BER=10 -3, a degradation of at least 2 dB
remains, compared to the limit bound. Results with GMMSE are not presented in
this figure because, as explained in Section V, they are exactly the same as those
of SU MMSE technique for a full load system.
We also note slight differences depending on the channels. As expected,
poorer performance is obtained over the VB channel - the most severe one. Over
the VA channel, for which results are not represented, we obtained the same
performance as over the HT channel. For the VB channel, we can note a floor
effect due to the time selectivity of the channel when OFDM duration increases;
in fact with only 256 subcarriers, which curves are not represented, there is no
floor effect. This phenomenon will be cancelled by channel coding.
VI.2.2. System performance versus the capacity
The performance of the SU MMSE, MU PIC and GMMSE detectors are compared for
systems with different number of users over BU channel. Figure 9 illustrates the
performance of each detector in term of capacity related to the number of active
users; the curves are drawn for a BER equal to 10-3 and consequently provide the
necessary Eb/N0 to reach this BER. As expected, performance improves when the
number of users Nu decreases since the MAI level decreases. As for the full load
system, MU techniques outperform SU techniques. The GMMSE curve tracks the
MU PIC curve fairly closely up to a 3/4 load system, then the PIC detector
outperforms the GMMSE detector. As mentioned in Section V, at full load, GMMSE
and SU MMSE have the same performance.
84
The total bit rate increases with the number of users NU.. Nevertheless, most
of time, systems are not full load but rather half load. Thus, in meantime, using
either a PIC detector or a GMMSE detector presented the same performance up to
48 users and they both outperform the SU MMSE detector. If implemented with an
adaptive algorithm, the GMMSE detector remains easier to implement than a MU
PIC detector for a lower complexity [25].
Figure 9. Performance versus capacity over a BU channel at BER -10
VI.3. Multiuser detector including channel coding

By introducing redundancy in the transmission system, channel encoding allows
better performance, especially when channel fading, noise or interference occurs
[37]. At the reception level, in order to achieve an optimal efficiency, the Viterbi
decoder requires a reliability measurement of the data given by a LogLikelihood Ratio (LLR) function. An estimation of the LLR function for MC-CDMA
systems using QPSK modulation is given in [38]:
where Li depends on the mean value of the channel fading over Nc subcarriers
on which the transmitted data has been spread and zi is the symbol obtained after
despreading.
Last results include either a classical Convolutional Coding (constraint
length K = 7, polynomial generators: 133o and 177o) or a Convolutional TurboCode (constraint length K = 5, polynomial generator: 23o and 35o). The two
encoders are half rate. As mentioned in [23], decoding might be introduced into
85
the IC process in order to obtain a better estimation of the interfering signals;

nevertheless the complexity largely increases with the number of active users,
limiting them to a small number. Thus, for our simulation, channel decoding is
performed outside the iterative process for GMMSE and MU detectors.
VI.3.1. Influence of time interleaving
Figure 10 illustrates BER performance with convolutional coding for a full load
system over HT channel. Curves respectively correspond to results obtained with
PIC and SU MMSE detectors with and without time interleaver. We implemented a
random 12 msec interleaver that corresponds to about twice the coherence time
of the HT, VA and VB channels.
For results obtained without time interleaver, we can note on the 2 upper
curves that PIC detector performance is worse than MMSE detector performance.
This phenomenon, already pointed out by [32], is mainly due to bad decisions in
the IC process which leads to bursts of errors at the output of the convolutional
decoder.
Figure 10. Influence of interleaving for a full load system

over a HT channel with convolutional code.
Thanks to the time diversity of the channel, time interleaving provides a
significant gain highlighted by results of Figure 10. Otherwise carrying out the
time interleaving allows PIC detector to outperform the SU MMSE detector at about
10-3 as shown by the lower curves in Figure 10. Introducing time interleaving
leads to a 3 dB gain for PIC detector and to a 2 dB gain for the MMSE detector at
BER = 10-4; these substantial gains confirm the opportunity of using a time
interleaver for these selective channels.
86
Figure 11 illustrates the performance of an MMSE detector over HT, VA and VB

channels with time interleaving for a full load system. We can note a 1.75 dB
gain on performance over HT channel at BER=10-3 when introducing time
interleaving, and about a 2 dB gain for channel VB. Over the VA channel that is
less frequency and time selective than HT and VB channels, the coding gain is
reduced to 1.5 dB.
Figure 11. BER performance with convolutional coding for a full load system,
different channels and MMSE detector.
VI.3.2. BER performance including a Turbo-Code for a half load system

Performance of a MC-CDMA system including a powerful Turbo Decoder [39]
is given in Figures 12 and 13. Results are obtained after 4 iterations of Turbo
decoding for both channels. Performances over VA channel are presented for a
half load system in Figure 12, where the limit bound, which takes into account
the Turbo Decoding process, is given as a reference. As illustrated by BER curves,
the MU PIC detector does not really outperform the SU MMSE detector. In fact,
reaching a 10-4 BER after decoding corresponds to a BER of about 5 x 10-2 before
decoding where performance of all detectors are equivalent (cf. Figure 8). Thus,
the interest of implementing either a SU or a MU IC technique will depend on their
respective performance before decoding [40].
Figure 12 also illustrates the good performance of the GMMSE detector
especially for a nonfull load system as already pointed out without channel
coding in Figure 9, where the GMMSE detector had equivalent performance to the
MU detector. Moreover, in the case of channel coding, the GMMSE algorithm
widely outperforms the MU detector. In fact, the PIC detector is not optimal at low
signal to noise ratios because of bad decisions taken in the estimation of the
interferers in the IC process.
87
Figure 12. BER performance for a half load system over a VA channel.
VI.3.3. BER performance with a Turbo Decoder versus the number of
active users
In Figure 13, performance of each detector over the BU channel is given versus
the number of users. As in Figure 9, the curves are drawn for a BER equal to 10-3
and consequently provide the necessary Eb/N0 to reach this BER.
For the BU channel, we did not carry out time interleaving. In fact, this
channel would require a 333 msec interleaver in order to fight symbol
correlation and implementing such a long interleaver is not possible for mobile
applications. Nevertheless, we can note that up to 1/3 load the GMMSE detector
allows performance close to the limit bound and is then always better than PIC
and MMSE detectors whilst the PIC detector never widely outperforms SU MMSE.
In fact, as already pointed by results in Figure 10, for a full load system, the
MMSE detector even outperforms the PIC detector. The interest of implementing a
Figure 13. Performance of a MC-CDMA system including a Turbo Decoder on a

BU channel.
88
GMMSE algorithm associated with a powerful Turbo Code is obviously

demonstrated here since it always widely outperforms the other detectors.
VII. Conclusion
In this paper, we have presented the different Multi Carrier Spread Spectrum
systems. We have described the MC-CDMA system well adapted to the downlink
of wireless high rate system cellular networks. We have detailed the existing
Single User and Multiple User algorithms and presented a new one called
GMMSE. We have demonstrated how efficient the use of the GMMSE detector was
over BU, HT and Vehicular channels, especially when associated with a powerful
Turbo Code. Even if some system parameters may be optimized depending on
the application, the performance of MC-CDMA systems presented in this paper are
quite acceptable leading to a total spectral efficiency (including the losses due to
the insertion of pilot symbols for channel estimation) of about 0.8 bit/(sec .Hz)
for each studied channel. This spectral efficiency corresponds to 0.0125 bit/(sec .Hz)
per user in our system where 64 users can simultaneously transmit. With respect
to UMTS and IMT2000 requirements, it is possible to allocate to one user several or
all user codes such that, for a bandwidth equal to 5 MHz, a net bit rate up to 4
Mbit/sec can be assigned to a single user. A 6 Mbit/sec bit rate could be reached
with a 3/4 rate code. Thus, it is shown that MC-CDMA is a very promising multiple
access scheme especially for the downlink of future mobile radio systems.
Nevertheless, we have to keep in mind that some parameters such as channel
estimation, nonlinearity amplifiers and phase noise oscillators, as well as
synchronization devices have not yet been taken into account to complete this
analysis.
Acknowledgments
The authors, J.-F. Hlard and J.-Y. Baudais from INSA, Rennes would like to
thank FT R&D/DMR/DDH and Mitsubishi ITE who supported and contributed to this
study. Furthermore, the authors would like to express their thanks to the
anonymous reviewer for his useful suggestions.
89
REFERENCES
[1] TURIN (G.L.), Introduction to spread-spectrum antimultipath techniques and their
application to urban digital radio, Proc. of the IEEE 68, n 3, pp. 328-353,
(March 1980).
[2] PICKHOLTZ (R.L), MILSTEIN (L.B.), SCHILLING (D.L.), Spread spectrum for mobile
communications, IEEE Trans. Vehic. Tech., 40, n 2, pp. 313-322, (May 1991).
[3] LEE (W.C.Y.), Overview of cellular CDMA, IEEE Trans. Vehic. Tech., 40, n 2,
pp. 291-302, (May 1991).
[4] SHAFI (M.), Wireless communication in the twenty-first century: a perspective,
Proceedings of the IEEE. 85, n 10, pp. 1622-1638, (1997).
[5] ALARD (M.), LASSALE (R.), Principles of modulation and channel coding for digital
broadcasting for mobile receivers, EBU review, 224, pp. 3-25, (Aug. 1987).
[6] LE FLOCH (B.), HELARD (J.F.), HALBERT (R.), LEMESLE (J.M.), Modulation et codage de
canal pour la radiodiffusion sonore numrique vers les mobiles, Colloque
GRETSI'89, Juan-Les-Pins, (12-16 June 1989).
[7] ETSI ETS 300 744, Digital video broadcasting (DVB): framing structure, channel
coding and modulation for digital terrestrial television (DVB-T). (March 1997).
[8] ETSI, Project Broadband Radio Access Networks (BRAN); HIPERLAN Type 2, Technical
specification; Physical layer, (Oct. 1999).
[9] IEEE, supplement to standard for telecommunications and information exchange
between systems - LAN/MAN specific requirements - Part 11: wireless MAC and PHY
specifications: high speed physical layer in the 5 GHz band, P802.11a/D7.0,
(July 1999).
[10] CHOW (J.S.), Tu (J.C.), CIOFFI (J.M.), A discrete multitone transceiver system for
HDSL applications, IEEE JSAC, 9, pp. 895-908, (Aug. 1991).
[11] YEE (N.), LINNARTZ (J.P.), FETTWEIS (G.), Multicarrier CDMA in indoor wireless radio
networks, Proceedings of IEEE PIMRC'93, pp. 109-113, Yokohama, Japan, (1993).
[12] FAZEL (K.), PAPKE (L.), On the performance of convolutionnally-coded CDMA/OFDM
for mobile communication system, Proceedings of IEEE PIMRC'93, pp. 468-472,
Yokohama, Japan, (1993).
[13] FAZEL (K.), Performance of CCDMA/OFDM for mobile communication system,
Proceedings of IEEE ICUPC'93, pp. 975-979, Ottawa, Canada, (1993).
[14] CHOULY (A.), BRAJAL (A.), JOURDAN (S.), Orthogonal multi-carrier techniques applied
to direct sequence spread spectrum CDMA system, Proceedings of IEEE
Globecom'93, pp. 1723-1728, Houston, USA, (1993).
[15] DASILVA (V.M.), SOUSA (E.S.), Performance of orthogonal CDMA codes for quasisynchronous communication systems, Proceedings of IEEEICUPC'93,pp. 995-999,
Ottawa, Canada, (1993).
90
[16] VANDENDORPE (L.), Multitone direct sequence CDMA system in an indoor wireless
environment, Proceedings of the first IEEE symposium of communications and
vehicular technology, pp. 411-418, Delft, The Netherlands, (1993).
[17] HARA (S.), PRASAD (R.), Overview of multicarrier CDMA, IEEE Communications
Magazine, pp. 126-133, (Dec. 1997).
[18] KAISER (S), OFDM-CDMA versus CDMA: performance evaluation for fading channels,
IEEE International Conference on Communications, pp. 1722-1725 (June 1995).
[19] OH (J-Y.), LIM (M-S.), The bandwidth efficiency increasing method of multicarrier
CDMA and its performance evaluation in comparison with DS-CDMA with Rake
receivers, Proceedings of IEEE Vehicular Technology Conference, pp. 561-565,
(1999).
[20] HELARD (J.-F.), BAUDAIS (J.-Y.), CITERNE (J.), Linear MMSE detection technique for
MC-CDMA, Electronics letters, 36, n 7, pp. 665-666, (March 2000).
[21] YEE (N.), LINNARTZ (J.P.), Wiener filtering of multi-carrier CDMA in a Rayleigh fading
channel, Proceedings of IEEE PIMRc'94, 4, pp. 1344-1347, (Sept. 1994).
[22] LE GOUABLE (R.), HELARD (M.), Performance of MC-CDMA Systems in Multipath
Indoor Environments - Comparison with COFDM-TDMA System, 3G Mobile
Communication Technologies, London, (March 2000).
[23] LE GOUABLE (R.), HELARD (M.), Performance of Single and Multi-User Detection
Techniques for a MC-CDMA System over Channel Model used for HIPERLAN2,
Proceedings of IsssTA2000, New-Jersey, USA, (Sept. 2000).
[24] HELARD (J.-F), BAUDAIS (J.-Y.), CITERNE (J.), Comparaisons des performances de
diffrentes Techniques de detection Appliques un signal AMRS Porteuses
Multiples dans un Canal de Rayleigh", GRETSI'99, Vannes, France, (Sept. 1999).
[25] BAUDAIS (J.-Y.), HELARD (J.-F.), CITERNE (J.), An improved linear MMSE detection
technique for Multi-Carrier CDMA systems: comparison and combination with
interference cancellation, European Transactions on Telecommunications, Special
issue on Multi-Carrier Spread-Spectrum, 11, n 6, pp. 547-554, (Dec. 2000).
[26] MOTTIER (D.), CASTELAIN (D.), A spreading sequence allocation procedure for
MC-CDMA transmission systems, Proceedings of the VTC2000, Boston,
pp. 1270-1275, (Sept. 2000).
[27] KAISER (S), KRZYMIEN (W.A.), Performance effects of the uplink asynchronism in a
spread spectrum multi-carrier multiple access system, European Transactions on
Telecommunications, Special issue on Multi-Carrier Spread-Spectrum, 10, n 4,
(July/Aug. 1999).
[28] DASILVA (V.), SOUSA (E.S.), Multicarrier orthogonal CDMA signals for quasisynchronous communication systems, IEEE Journal on Selected Areas in
Communications, 12, pp. 842-852, (June 1994).
[29] SOUROUR (E.), NAKAGAWA (M.), Performance of orthogonal multicarrier CDMA in a
multipath fading channel, IEEE Trans. on Communications, 44, n 3, pp. 356-367,
(March 1996).
91
[30] KONDO (S.), MILSTEIN (L.B.), Performance of multicarrier DS-CDMA systems, IEEE Trans.
on Communications, 44, n 2, pp. 238-246, (Feb. 1996).
[31] PROAKIS (J.G.), Digital Communication, McGraw Hill, 1995.
[32] CORNIC (A.C.), GRANT (P.M.), THOMSON (J.S.), A Novel Interference Cancellation
Receiver for Uplink Multi-Carrier CDMA, Proceeding of ISSSTA 2000, Parsippany NJ
USA, pp. 155-158, (Sept. 2000).
[33] PATEL (P.) and HOLTZMAN (J.), Performance comparison of a DS-CDMA system using a
successive interference cancellation (Ic) scheme and a parallel IC scheme under
fading, Proceedings of IEEE Icc'94, New-Orleans, USA, (May 1994).
[34] BAUDAIS (J.-Y.), HELARD (J.-F.), Procd d'galisation dans des rcepteurs utilisant
une combinaison des techniques de modulations porteuses multiples et accs
multiple par repartition de codes. French patent n 99/11689, (Sept. 1999) and
n 99/15919, (Dec. 1999).
[35] COST 207, Digital land mobile radio communications, tech. rep., Luxembourg:
Office for Official Publications of the European communities.
[36] UMTS standard TR 101 112 V3.2.0, pp. 39-43, (April 1998).
[37] MAXEY (J.J.), ORMONDROYD (R.F.). Multi-carrier CDMA using convolutional coding
and interference cancellation over fading channels. First Workshop on MultiCarrier Spread-Spectrum, Oberpfaffenhofen, Germany, pp. 89-96, (April 1997).
[38] KAISER (S.), Multi-Carrier CDMA Mobile Radio Systems - Analysis and Optimization
of Detection, Decoding and Channel Estimation, PhD thesis, University of
Munich, Germany, (1997).
[39] BERROU (C.), GLAVIEUX (A.), THITIMAJSHIMA (P.), Near Shannon limit error-correcting
coding and decoding: turbo codes, Icc'93, Geneva, (1993).
[40] LE GOUABLE (R.), HELARD (M.), Performance of a MC-CDMA system including Turbo
Codefor wireless indoor communications. ECWT Paris, (Sept. 2000).
Chapter 4
Interpretations and performances

of linear reception in downlink
TD-CDMA and multi-sensor
extensions
L. Ros and G. Jourdain
Images and Signals Laboratory, Domaine University, France
M. Arndt
France Telecom R&D
I. Introduction
The normalization group 3GPP is preparing the standards for the third generation
of radio-mobile system, UMTS (Universal Mobile Telecommunications System).
Among them, the TDD (Time Division Duplex) mode [21] permits, for one or
several users, the simultaneous transmission, during each TDMA slot, of spread
QPSK symbols with different CDMA signatures. The uplink and downlink work at
the same carrier frequency around 2 GHz, with quadrature modulation by two
3.84 Mchip/sec-binary streams of 5 MHz-spectral width. A synchronization
channel and a midamble, transmitted at the slot rate, make the synchronization
and channel estimation easier, independently of the symbol estimation mainly
considered here. The spread spectrum systems were initially introduced in a
single-user situation in order to ensure discretion or to fight against undesired
channel effects [2]. They were calibrated, as far as possible, so that the symbol
time was superior to the temporal spread of the channel, in such a way that the
receiver was simply based on a matched filtering (MF).
Therefore, the transmission was robust but featured a very bad spectral
efficiency.
93
In contrast, the systems which are considered today should provide an increase
in the network global capacity. So, the spread messages from different users are
superimposed1, in a smaller number but as close as possible as to the spreading
factor; this results in a non-negligible part of Multiple Access Interference (MAI).
Moreover, the channel spread may go beyond the symbol time in the vehicular
environment [20], leading to substantial Inter-Symbol Interference (ISI). To
compensate interference and better detect each user's symbol, linear joint detection
with temporal depth is a receiving strategy which uses the knowledge of the codes
of other users, whereas the simple matched filter considers the MAI and IsI simply
as wideband noise. The linear joint detection is not optimal for minimizing the Bit
Error Rate (BER) but ensures a good compromise between performance and
complexity, especially in TDD mode where the codes are short (16-chip long
maximum). Nevertheless, interference cancelling, complete or partial, leads to an
amplification of thermal additive noise. This effect can be largely reduced by using
multisensor coherent processing on the mobile, which improves the waveform
orthogonality and facilitates the system inversion. We only consider non-adaptive
antenna for the base station transmitter.
The objectives of this chapter are:
(A) to interpret the CDMA system in downlink (in Section IV), usually with a
frequency approach, natural in spread spectrum systems. We will establish the
wide-band equivalent model, leading to an analogy with source separation in
antenna processing, and the symbol-band equivalent model which is identical to
the one not featuring spreading systems,
(B) to make a synthesis of the theoretical structures and performances of
"classical" linear receivers for selective channels (in Sections III, V and VI),
established from models described in (A). An exhaustive synthesis has been carried
out by the authors of [4] but from block temporal formulation using a transfer
matrix of all the symbols in the slot. The frequency approach permits here
simplification of the model and to take into account the selective channels naturally,
(C) to evaluate the average performances in Vehicular and Indoor
environment models (in Section VII) by application of the expressions of (B) in
order to measure the interest of joint detection and multi-sensor reception over
selective and non-selective channels.
II. Multi-user transmission model

Continuous-time baseband representation (complex envelope) of the received
signal, observed during one time slot (containing M symbols), is modelled as:
1. To simplify the description we abusively associate, in the whole article, one active code to one
user. In real applications, several codes may be affected to the same user in order to increase its
information rate (multi-code situation).
94
Interpretations and performances of linear reception
Figure 1. Baseband model of transmission.
where
- ak[.] are the QPSK symbols transmitted by the k-th source at the "symbol
times" mTs,
- gk = ck * he * h is the overall wide-band waveform of the k-th user.
It results from the convolution between the k-th spreading code, the half
Nyquist (square-root raised cosine filter "RRC" with roll-off r off ), and the
channel,
- n(t) is a baseband additive complex noise.
We consider the following context of downlink multi-user communication:
- the number of users (or active codes) K is less or equal to the spreading
factor Q;
- the K active codes (taken among a set of Q known codes) are supposed
known at the receiver, which is not necessarily the common situation
(necessity of a control channel);
- the codes of different users are made from Q binary complex entity named
"chips". The codes are linearly independent and orthogonal at zero delay
(algebraically meaning), which is expressed by:
where 5 is the Kronecker symbol whereas 5(.) is the Dirac distribution. The
impulse response of one code lasts for Ts, and is defined for the k-th user
by:
ck() =
Q-1
q=0
ck[q]
- qTc), where Tc = Ts/Q is the chip duration;
- the transmitted symbols are stationary, with zero mean and power A2; they
are uncorrelated temporally and from one user to another, and also
uncorrelated with the additive noise;
Advances in
UMTS technology
95
downlink with non-intelligent Base Station: all K users signals appear

at each mobile sensor after transmission over the same propagation channel
and with the same power. The channel h is supposed to be deterministic,
time-invariant during the observation (1 slot), linear and causal with a
temporal spread limited to W, chips. The structures studied at the sequel
need only these assumptions for the channel. Nevertheless in radio-mobile
communications, the channel is generally constituted by multiple paths, and
its impulse response is modelled by:
where L, is the number of paths, aiare the complex amplitudes and tithe
propagation delays of different paths, with zi s W, Tc,Vi{ 1, ..., Lt};
-the complex noise is an additive white Gaussian noise, with two-sided
NO for the
power spectral density 2N0 (corresponding to a two-sided psd 2
real noise on band carried at the entrance of the receiver);
- two slots of symbols are isolated by a guard time above W, * Tc ;
- we suppose perfect timing recovery (using the synchronization channel)
and also perfect channel estimation (using the midamble).

We suppose by convention only one desired code (or desired user), the code
number one.
We define the average energy per bit at the entrance of the receiver
(on band carried ) for the desired user:
Eb = A2Ts y,l,ol, where yllll is the zero-delay auto-correlation of the

4
Ts
waveform g , ( r ) of user 1.
Note: the assumption of no variation of the channel during the slot
(T,,, = 667 p e c ) is fully justified for a mobile moving with a speed w, of
3 W h . Indeed, for a carrier radio-wave (celerity wo) around f o = 2 GHz,
the Doppler spread is defined by Af
- 3 . f o and the channel coherence
d -
uo
=Afd- is around 200 msec. The hypothesis limit occurs with

time Tcoh
vehicular moving at urn = 120 k d h (Tcoh- 5 msec) since the complex
amplitudes of the paths may suffer from phase rotation up to
~~
1
2 Ti,
Tcoh
. 360 around *25 during one 112 slot, in the case of a deterministic
Doppler model. We might need to update the knowledge of the complex

amplitudes several times in the slot for optimizing the symbol detection.
96
Figure 2 . Joint linear detector (with ylllol = I ) .
On the other hand, since the mobile speed v m is always negligible compared to
the wave celerity q,,the variation of the path delay % . Tsroris negligible during
IJO
the slot (around 10-4Tc).
111. Structure of the joint linear detection

With infinite response, the symbol by symbol linear detector [17] can take the
shape of an imposed structure with one matched filter bank followed by a bank
of discrete equalizing filters working at symbol rate, as seen in Figure 2. It is
easy to verify this, under classical criteria (Zero-Forcing, MMSE).
111.1. Channel- and code-matched filter bank

The receiver front-end applies to r(t) a bank of continuous filters gkH,k = 1, ..., K
matched to each global waveform, made by the chip pulse (RRC), the channel
response and each of the K codes. By convention, the exponent ( . ) H represents
hermitian transform for a function h (i.e. hH(t) = h* (- z): matched filter) just
as the hermitian transposition for vectors or matrix. The K outputs of filters gkH
sampled synchronously at symbol time, constitute a sufficient statistic of the
continuous signal received r ( t ) for the symbol estimation [ l , 161:
Thus, knowledge of (Yk,ml},

k = 1, ..., K, m = 1, ...,M+ is sufficient to know the
likelihood function of the received signal and to carry out the optimal processing
which minimizes the error probability (or BER). For the treatment of one slot, we
must consider a number M+ of samples slightly greater than the effective number
M of symbols because of the channel temporal spread and because of the
matched filter. In this chapter, we will not consider the optimal theoretical postprocessing, i.e. the Maximum Likelihood Sequence Estimator, but only the
detectors which treat linearly (Ykcrn1}and make the decisions symbol by symbol.
97
The Ts-sampled outputs {yk[m]-} are expressed from discrete time convolutions of
the transmitted symbols, disturbed by an additive noise:
where ik [n] is the sampling, at symbol time, of the continnous and deterministic
cont
H
cross-correlation function ik() = (gi * g k) () between the wide - band
waveforms of users i and k:
After matched filtering and synchronous sampling, equation (5) shows that
all the system can be modelled at discrete time Ts. The series {yik [n]}, Vn Z
providing the discrete filtering, has no dimension. Through a multi-path channel,
the waveforms are generally not orthogonal and some ISI and MAI interference
occurs with secondary terms 11 at non-zero delays and cross-channel terms 1i,
i 1 at the different delays respectively:
The necessity and the difficulty of the equalization will be entirely conditioned
by the values {yik[n]}.
Recalling that:
* In the case of a single-user system or without joint detection, only the first
branch is used, operating a coherent combination relative to the different paths,
called Rake; y1[m] may then be the decision variable.
* In the case of a multi-user system but a single-path channel, the duration
of the code limited to Ts (and respect of the Nyquist criterion at the chip time for
he) leads to no interference at non-zero delay since ik[n] = 0, Vn 0. Moreover,
when the codes are orthogonal, there is no interference at zero delay since
Yik[0] = ik: only the first branch of the MF bank is useful, reduced to the receiving
filter RRC followed by correlation with the desired code.
98
* In the case of a multi-path channel with weak temporal spread relative to

symbol time, the contribution at non-zero symbol time delays are negligible and
only the MAI at the zero delay may be considered. It is similar to a situation
where the waveforms would last 7s, but without being orthogonal: this is the
case studied by most authors [16, 6, 5], corresponding to a well calibrated
system, but it leaves out a number of situations in a vehicular environment with
the TDD-UMTS parameters.
III.2. Bank of discrete equalizers (Ts)

In order to form a decision variable d1[m] used to estimate the user's one QPSK
symbols a1[m] by quadrant detection, an equalizer linearly combines the outputs
{yk[m]} of the K branches of the matched filter-sampler bank, after passing them
through a bank of discrete filters {e 1k[n] } working at symbol time:
Having described the structure of the detector, we will now derive and
interpret equivalent models of the CDMA system before setting the performances
and expressions of the multi-user equalizer.
IV. Interpretations and models of CDMA

During transmission, equation (1) expresses the multiple access based on
orthogonal waveforms (symbol pulse-shapes) from one user to another, for an
ideal channel. For the particular CDMA case, each pulse gk, for k = 1 to K, occupies
all the available time-frequency plane (Ts Q\Ts)instead of being located in time
(widthTs\Kin TDMA) or in frequency (width1\Tsin FDMA). The number (Q-K\Q) plays
the role of a code margin with regard to system saturation, same as the margins
provided by the guard times or the guard bands in TDMA or FDMA.
By introducing in (10) the impulsive form ak(t) of the different user symbols,
the signal received defined in (1) can be expressed classically as a time
continuous convolution between the symbols and the global wide-band
waveforms, disturbed by an additive noise:
hwit
99
This expression allows describtion of the chain as a cascade of linear filters

excited by the impulsive symbols streams, exactly the same way as a linear
modulation without spreading. The difference here is that the pulse shape is
wide-band compared to the symbol band of width1\Ts.This formulation will
facilitate the frequency transformation in the following paragraph.
I V.1. Interpretation at transmission (multi-band transmission)

The result of a single observation during the slot can be analysed in the
frequency domain with the Fourier transform. The signals a(t) and n(t) are
considered here as deterministic (being particular occurences during a slot of
random variables), which avoids treatment with (cyclic) autocorrelation
functions of the received signal. The Fourier Transform of equation (9) gives:
where by convention:
-fw is used for any frequency while / is a frequency of the symbol-band,
included in [1\2Ts,+1\2Ts].So, for all fw there is a single frequency / and
a single relative integer q such that fw = f + q\Ts,
- the continuous Fourier transform of a function h in continuous time or
discrete time (then obtained via Z transform) is noted h(.).
The transmission bandwidth, defined by the support gk(fw) is limited to
the band of the half Nyquist filter, that is Qr.1\Tswhere Qr is the integer part of
(1 + roff ). Q, expressing the filter bandwidth excess. With the UMTS parameters
Q = 16 and roff = 0.22, Qr is odd and equal to 19.
The Qr symbol bands (of width1\Ts)which comprise the wide-band are then
centred around the frequencies fw = Qr/2\Ts,..., 0, ..., +Qr/2\Tswith the definition of
Qr/2 such that Qr = 2Qr/2 + 1.
Note: Taking into account the RRC attenuation, the portions not included in
this Qr bands are ignored. If one does not want to lose any information, we might
have taken Qr + 1 = 20 instead of Qr and choose for f a range equal to [0, + ].
Due to the characteristic of "sampling" (cf (10)) of ak(t) at the symbol rate
(or the cyclostationarity of the digital communication signals), k(fw) is periodic
with a period of 1\Ts:
100 Interpretations and performances of linear reception
Expression (11) is then written inside the useful wide-band:
with Vq = -QT/2,..., + QT/2 :
Equation (14) expresses the frequency diversity of order Qr inherent to

spread spectrum systems: for a given user "k", the same binary message ak[m] , m
= 1, ..., M, with Fourier transform k (f) on the complete symbol band, is
transmitted in Q`r different symbol-bands. Generally (with weak excess
bandwidth or roll-off), the portions outside the chipband (width 1\Tc) are
attenuated and only a diversity of order Q is considered, according to the
temporal structure of codes, measured at the chip rate.
In an equivalent way, emphasis can be put on diversity by means of a multicarrier modulation concept. From the periodicity of k (f) and the impulsive
form of the code ck() with finite length Ts, the k-th user's signal, before RRC and
channel filtering, is expressed in frequency domain by:
Unlike OFDM (Orthogonal Frequency Division Multiplexing), for a given

user, the sub-carriers are modulated by the same information (complex binary
message). There are spaced by and have complex amplitudes ck(-q\Ts) given by
the DFT (Discrete Fourier Transform) of the Q chips of the code. The messages
of different users will easily be separated at the receiver, at least over ideal
channel, thanks to the codes orthogonality (independency) property, which is
expressed in the frequency-domain by the linear orthogonality (independency)
of the different sets of Q sub-carriers {ck(q\Ts)}.
Note: we can establish equation (15) more naturally by representing the
spread-spectrum as the result of the multiplication of the continuous symbol
Advances in UMTS technology 101
stream akrect(t) (shaped into a rectangular pulse of length Ts) by the periodic
version of the code ckper(t) (with period Ts). The expression of the periodic code
in Fourier series gives then the multi-carrier modulation scheme directly. This
other equivalent representation of the spreading system, i.e. the "multiplicative
representation", is closer to the design, and probably more traditional [3].
However, it is more difficult to manipulate theoretically than the representation
with linear filters, generally adopted.
IV.2. Interpretation at reception
Assuming a perfect knowledge of the channel and timing synchronization, the
outputs of the bank of MF-samplers can be synthesized from a linear
transformation of the transmitted symbols, without mentioning the wide-band
received signal. The temporal expression, at discrete time Ts, of the "symbolband model" has been established in (5).
The transfer yik(f) between the symbols of the source "i" and the output of
the branch "k" is expressed in frequency-domain by Fourier transform of
equation (6) as the aliasing of the wide-band cross-spectrum y ik c nt (f w ) in the
symbol-band:
with:
We can now explain the notions of aliased equivalent channels in the

particular case of CDMA.
IV.2.1. Aliased Tx/Rx global channels
At the first branch output, the transfer y 1 1 ( f ) is equivalent to an aliased TransReceiving Global Channel (aT/RGC) in the symbol band for the desired user, the
transfer i1(f) to an aT/RGC for the interfering user "i 1". Expressed in
frequency, the absence of IsI on the branch "1" corresponds to an aT/RGC white
for the desired user, and the absence of MAI corresponds to the nullity of the
aT/RGC for all the interfering users. These properties are absolutely false for the
T/RGC before aliasing, Yikcont (fw), because of inter-chip and inter sub-chip
interference on the wide-band channels outputs before sampling. It simply
expresses, for the desired user, a coherent recombination of the sub-carriers and
Figure 3. Tx/Rx Global Channel example in VB.
for the interfering users, a destructive re-combining. With a single-path channel,

the orthogonality (2) and the finite length of the codes ensures these conditions.
It is easy to show using the Poisson formula that Vf[ 1\2Ts, +1\2Ts]:
Respect of the first Nyquist criterion, relative to the chip time Tc for he * he H,
leads then to:
For the desired user, equation (19) simply expresses the first Nyquist
criterion relative to the symbol time Ts for g1 * g1 H with an ideal channel. When
the channel is not single-path, the aliasing equation (16) explains the robustness
of the spread-spectrum to selective channels: the aT/RGC seen by the desired user
symbols is finally a channel with width obtained from the coherent
75
superposition of Qr channels with equal width, averaging the selectivity imposed
to the different bands.
Figure 3 shows the T/RGC and aT/RGC on branch one for the desired user and
for the interferer "2", with a channel taken from the Vehicular B model,
characterized by a temporal spreading around 5 Ts. The power of Interference

and Noise (IN) at the first branch output is derived from the temporal expression
of the ISI, MAI and Noise terms defined in (7):
The expression of (IN) vs frequency, established with the Parseval relation

applied to equation (20), shows that the Interferference and Noise power
depends only on the form of the aT/RGC:
The power of IsI is derived from the quadratic frequency area between the
desired aT/RGC and an ideal flat (white) one: it results thus from the
non-whiteness of the desired aT/RGC. The power of MAI is derived from the sum
of the quadratic frequency area of interfering aT/RGC relative to the branch "1".
The noise power is derived from the Noise Equivalent Bandwidth: Y11[10] . 1\Ts of
the branch "1", i.e. from the frequency area in the symbol band of the desired
aT/RGC.
IV.3. Equivalent matrix models

For compacting and simplifying the representation of the downlink CDMA
system, we can use two equivalent matrix models, named wide-band model and
symbol-band model. The diversity can well be represented by the wide-band
model, by showing the Qr bands containing the signal received. It may be useful
when some parameters (channel, codes, timing synchronization) are not
identified. Otherwise, being the context of this article, the representation model
in the symbol band will be sufficient to establish the expressions and
performances of the equalizers.
IV.3.1. Wide-band model

The Qr components of the signal received at frequency f in the different bands
(equations 13 and 14) are expressed algebraically by:
where
The matrix G(f) of size Qr X K contains the wide-band waveforms relative

to f in the different symbol-bands, and for the different users. It plays the role of
a transfer matrix from the K sources towards the Qr bands, for a given frequency
of the symbol-band.
An analogy can be made with the source separation from the reception on an
array of Qr sensors (in place of Qr frequency bands). In antenna processing
domain [19, 8], each of the Qr sensors receives a mixture of K sources but the
latter arrive from different directional (spatial) vectors. In CDMA, each of the Qr
symbol-bands of the reception contains a mixture of the K sources (each band
containing all the information for a given source) but the sources "arrive" from
different "frequency vectors", imposed by the codes and the channel.
We find again some features which are also classical assumptions in antenna
processing:
- the noise is uncorrelated from one band to another:
- the uncorrelation between the sources leads to:
where IN denotes the identity matrix of size N X N.
The beamforming favours one spatial direction for extracting the source "1"
from the coherent recombining of the outputs of the Qr sensors. In the same way
in CDMA, the matched filter favours one "frequency signature" by coherently (in
branch "1" for the desired user) re-combining the components of the Qr bands
or sub-carriers, which appears in (4) and can be formulated as:
where
The equivalent diagram of transmission and reception for the wide-band

model is given in Figure 4:
Figure 4. Transmission multi-band model.
IV.3.2. Symbol-band model

Spectral matrix of the aliased cross-channels
We define, for each frequency f belonging to [- 1/2 Ts; + 1/2 Ts], the normalized
spectral matrix of the cross-channels aliased in the symbol-band. This matrix
groups all the AT/RGC in the following form:
This matrix is of a K X K size, hermitian and non-negative definite. It will

be with full rank and so positive definite for all f if the spectrum | h (fw) | 2 of
the propagation channel does not have zero in the wide-band. Note that the
absence of zero is a sufficient but not a necessary condition. On the contrary, if
the spectrum of the propagation channel has a common zero at a same
frequency / in more than (Qr K) different symbol-bands, the matrix is
necessarily with rank deficient for this specific frequency. Therefore, except for
pathological channels, the normal rank (i.e. for almost all f) will be full, equal
to K.
In terms of interference for the whole users, ISI corresponds to the "nonwhiteness" of the diagonal terms of (f), and MAI to the non-zero non-diagonal
terms.
The formulation of (f) from G(f) is expressed in a matrix form from the
aliasing relation (16):
Matrix form of the model

The symbol-band model (Figure 5) links directly the K sources of symbols to the
outputs of the K branches via the transfer matrix ( f ) . It can be derived from
Figure 4 (i.e. equation (22)) of the wide-band model or directly from equation
(5) after passage through frequency and matrix formulation:
where
Figure 5. Symbol-band model of the transmission.
Figure 6. Matrix / ( f ) / example in Pedestrian B.
The equivalent discrete noise k[m], artificially reported in addition on each

branch ouptut is gaussian, correlated temporally and from one branch to another.
Its cross-spectral density, from branch "i" to branch "k" is
Two examples of the matrix | (f) | are given for K = 8 users (with linear
scales): in Figure 6, the temporal spreading of the channel (one realization of the
Pedestrian B channel according to the ETSI model) is around 1 75.
In Figure 7, the temporal spreading is around 5 Ts (same Vehicular B channel
as used in Figure 3). In the two examples, the interference is around the same
order but we can verify that the coherence bandwidth in Vehicular B is about 5
times smaller.
Figure 7. Matrix | ( f ) | example in Vehicular B.
V. Expressions and performances of linear detections

V.1. Matched filter only
V.1.1 Signal to interference and noise ratio
The matched filter has the feature of maximizing the Signal to Noise Ratio (SNR)
at mTs times, without considering the interference (isi and MAI). To take the
interference into account, the quality criterion of the Signal to Interference and
Noise Ratio (SINR) is best appropriate.
For the desired user ("1"), the SINR in output of the first branch of the
matched filter bank is expressed in the matrix (f) by:
Advances in
where 1'
2
-
UMTS
technology 109
[ l , ..., l],LT! [1,0, ..., 01
In the absence of ISI and MAI, the matched filter obtains the best
2E
to (>)
as well as the ideal elementary Binary Error Probability.
NO
SINR
equal
V.2. Equalizers: ZF and MMSE
In the presence of interference, a numerical linear equalizer bank running at Ts

completes the head of the base-band receiver, as set forth in Section 111.2.
The decision variable d,,,] defined in (8) can be written as a global
convolutive discrete transfer {gloli[,],i = 1, ..., K ) for the symbols, disturbed

by an additive discrete noise b,:
The Mean Squared Error (MSE) is generally used to describe the

performances of the equalizer. This energetic measure benefits from a simple
quadratic formulation and presents simple solutions of minimization problems.
The MSE in the output of the equalizer for the desired user is expressed in
frequency terms by:
(MSE>& E{ 1 d,,,]
a,[,] 1 2>
The integration term in this expression is the power spectral density of the
error (psd), comprised of a ISI term, a MAI term, and a noise term.
From matrix formulation:
- the vector t?, (f) = [e^,,cf), ..., gIKcf)] is created from the transfer
functions ofthe equalizing filters on the K branches,
-
the global transfer function for the symbols of the user k is: giolkcf)= e^,-
<ne E 0 - k
wherekT = [..., 0, 1, 0, . ..I

1
k
- the global transfer function t1(f) for the noise (with periodicity due to
the sampling) is such that:
The psd of bl[m] is then 2N0 |t1(f) | .

From these expressions, it is easy to establish, by omitting (f) in the second
member for the sake of clarity, that:
Notes:
* Interference power plus noise can be obtained simply from the MSE by
subtracting the quadratic error term at zero delay: (IN) E q = (MSE) E q
A2 \ l-glo 1 1 [ 0 ] \2.
* The global linear detector 11 for the detection of the symbols of user "1",
including the MF-sampler bank and the equalizing filter bank, has a transfer
function l1(fw) defined for the wide-band. It achieves actually, for one frequency
f of the symbol-band, a linear combination of the Qr sub-bands. Its expression
under vector form of size Qr (grouping l 1 ( f - Qn/2\Ts).... l1(f + Qn/2\Ts), is:
V.2.1. ZF equalizer
The optimal ZF equalizer minimizes the MSE, under the constraint of cancelling
the ISI and MAI. With such a criterion, in situations without noise, the estimated
symbols must be equal to the transmitted symbols for the desired user. At the
same time, among all the wide-band global filters l1 which cancel the
interference, this criterion warrants the one with a minimal norm and leads to the
structure in Figure 2 (or equation (27)). The cancelling of the interference from
the symbol-band model requires then that glolk(f) = k1, which is equivalent to
glo1k[n] = k1n in discrete time representation.
The equalizer vector is then obtained by inverting the spectral matrix of
aliased cross-channels:
This is immediately derived from the equation (23) of the symbol-band

model. So, the resulting MSE is formed only by the contribution of the noise,
amplified by the interference cancelling process:
The zero delay coefficient of the equalizer in branch "1", e11zf[0], is a posi
real number, derived from the diagonals of the Hermitian matrix. It plays a
fundamental role since it describes entirely the performance of the system. More
precisely, it expresses a degradation factor for the SINR but also for the Error
Probability (Pe or BER)2 since after perfect theoretical equalization, we obtain
exactly the desired symbols, only disturbed by a gaussian noise:
The degradation factor corresponds to the frequency-domain area of the first

diagonal "block" (or frequency pattern) of the inverted aliased cross-spectral
matrix:
The Schwartz inequality lets us verify that e11zf[0] 1 . Of course, no

degradation (e11zf[0] = 1) corresponds to an ideal situation with a single path
channel and orthogonal codes.
Figure 8 gives the example of an inverted matrix | (f) - 1| in Pedestrian B.
The patterns of the first line represent the frequency modulus of the equalizing
filters 1k for k = 1 to 8; the degradation factor for the desired user "1" is rather
weak, computed to +1.2 dB.
Note:
From the received signal r(t), there are an infinity of linear detectors l1 able to
cancel interference. The linear ZF equalizer recalled here (with theoretical
frequency description and therefore no constraint of finite length for the impulse
responses) is optimal since it cancels interference but also minimizes, under this
constraint, the noise amplification. It is not the case for a ZF detector which
would consist in inverting the wide-band continuous channel, in the bandwidth
of the RRC, before making the correlation with the desired code. Such a detector
would generally present much worse performances.
Figure 8. Matrix / (f)
- 1
/ example in Pedestrian B.
V.2.2. MMSE equalizer

The mean squared error is at a minimum when the quadratic form (26) of the
spectral density of the error is minimized for each frequency / of the symbol
band, which leads to:
with T'(f) = T(f) + (N0/2F ) I=K

The expression for the MMSE is then:
The inversion of the aliased cross-channels spectral matrix is realized here

after adding a noise to signal N0/2E factor on the diagonal. So, as in a single user
2Eb
situation, the partial inversion of the MMSE limits the amplification of the noise to
the frequencies where the local signal to noise ratio is bad. Moreover, since the
power of the noise is not zero, the inverse of '=(f) always exists an ill conditioned
situation is impossible, contrary to the ZF case.
The global gain (at zero delay) of all the chain Tx/Rx for user "1 ",glo11me[0]
=
Ts f 1_ [T'(f)-1 T(f)]11df, is a positive real less or equal to 1 (equal to 1 in
a noise-free situation). The MMSE receiver has the same behavior as the ZF
Eb
receiver in situations of strongEb/N0and is reduced to the receiving head, i.e. to
the MF, inthe extreme opposite situation. The zero delay coefficient of the
equalizer is here normalized by the global gain to form the degradation factor
of the SINR ( 1), which is expressed through:
For strongEb/N0situations, the SINR of the three receivers described in this

0
chapter are ordered in the following manner:
Then, the performances of the receiver are expected to be improved by

equalization, preferentially MMSE if the noise level is not negligible.
V.2.3. Comparisons and design elements
The multi-user equalizer we have just described generalizes the classical results
obtained with weakly selective channels, based on the correlation matrix of the
codes at zero delay. When the temporal spread of the channel is weak compared
to the symbol duration, the equalizer needs no memory dependence (white
response in frequency) and is reduced to instantaneously and linearly combining
the yk[m]. In this case, the spectral matrix of aliased cross-channels, white in
frequency, but non-diagonal if the coded waveforms are not orthogonal, is
identical to the zero delay correlation matrix, as the frequency dimension does
not appear. We then find the equalizer described by the authors of [16, 5, 6],
where the coefficients combining e1k[0], from the branches k = 1 to K are given
by the first line of the inverse of the zero-delay correlation matrix of the codes,
equal to [1,0, ..., 0] if the codes are orthogonal.
In the presence of selective aT/RGC in the symbol band, the equalizing filters
may use practically a finite temporal depth of P coefficients (with causal and
anti-causal parts). We then have added the frequency dimension /, which is
equivalent to independently treating several non-selective channels, with narrow
bands compared to (second order stationarity).
This approach naturally (but approximately) provides the coefficients of the
equalizer by using the Inverse Discrete Fourier Transform, and constitutes an
alternative solution to temporal methods [12, 4], with little complexity.
To obtain a temporal depth of P coefficients for the equalizer, we just have
to sample the symbol band with a step f =1/PTsto form the P matrix=(p f)
and compute their inverse. The K non-causal filters {e l k [ p ] } used to detect the
symbols of user "1" are obtained by a DFT - 1 of the K spectral patterns of the
line "1" of the inverse matrix. The filter {e 1 1 [ p ] } born from the diagonal is the
most energetic and corresponds to the main branch. In a multi-code case, i.e.
with several desired codes, we can take advantage of the hermitian symmetry of
the taps: eik[n} = e*ki[_n], Vi, k = 1...K, Vn ZZ.
Notes:
* The impulse responses of the equalizer bank at symbol time after the MF
bank are theoretically infinite. So, the choice of P should ensure a negligible
temporal aliasing for the practical filters. Depending on to the weak values of the
secondary far paths in the ETSI models described in Section VII, a typical
temporal depth of 1 to 4 coefficients in indoor A and of 8 to 16 coefficients in
Vehicular B is generally sufficient.
* Instead of preserving approximately the theoretical imposed structure
described in Figure 2 (with a finite number of taps for the practical equalizer), it
is also possible to compact the linear detector l1 in a single discrete transverse
fractional filter with a cadence of Tc/2 at input and Ts at output.
VI. Extension to multi-sensor reception

When the mobile receiver uses L = 2 or 3 sensors (with a spacing above /2 =
7.5 cm), the i = 1,..., Lt propagation paths are observed on the different elements
/ = 1,..., L of the antenna. The delays of one path "i" on the different sensors "l"
are nearly identical according to chip time (cf note in Section II), only the
complex amplitudes ali are different. Thus, we define L propagation channels
h(l)() from the model of (3), still considered here as deterministic and time
invariant during the slot.
With the l-th sensors are also associated a received signal r(l) (t), a spatiotemporally white noise n(l) (t), a wide-band waveform for the k-th user gk(l)() =
and transfer matrix sources/bands G= ( l ) (f). With multi-sensor,
and Eb represent respectively the normalized "energy" of the waveform

and the average energy per bit of the desired signal over the L receiving sensors.
The channel matched filter becomes spatio-temporal [1, 7, 11, 15] but similar
to the multi-band dimension, the spatial dimension does not appear explicitly in
the "symbol-band" model and the digital equalizer keeps exactly the same
expressions as function of T(f) rather than those established in single-sensor,
described by equations (28, 29, 30, 31, 32). Obviously, the global diversity
2
(l)
2
is concealed in =(f) since | h \ is replaced by | h^ | in expresion (17) of
cont
the T/RGC ^ i k (f) and thus indirectly in the formation of the a T/RGC^ik(f),
leading to:
The coherent re-combination of the spatial and frequency channels improves

the orthogonality of the waveforms and therefore makes easier the system
inversion operated by the joint detection equalizer. In terms of implementation,
we can approximate (with finite length) the imposed structure of Figure 9, taking
benefit of the discrete nature of the paths in the channel MF. Or we can
implement linear receivers by means of one single fractional filtering per sensor.
In the next section we are now going to measure the benefit of the multi-sensor
reception.
Figure 9. Multi-sensor transmission model.
VII. Average performances in indoor and vehicular

environments
VII.1. Context
VII.1.1 Codes used
According to the general hypothesis, the codes {ck [q]q = 0, ..., Q 1} for the
users k = 1, ..., K are complex with values proportional to {1, j, 1, j},
algebraically independent and moreover, in conformity with the norm [21],
algebraically orthogonal (cf (2)). One code is comprised of Q = 16 chips
obtained from a "channel code" {chk[q],, q = 0, ..., Q 1} and a "scrambling
code",
= 0, ..., Q 1}. The channel code is particular to one user, with
real binary values ( { 1; + 1}) chosen in the family of 16 orthogonal WalshHadamard sequences. The scrambling code is specific to each cell, with real
binary values indexed in the norm. The complex values of the final code are
obtained by a rotation of /2 from one chip to another after multiplication of the
two previous codes:
where j is the an imaginary number such that (j)2 = 1.

So, with such a complex code, both real and imaginary parts of the
transmitted symbols benefit by the diversity of the In phase and Quadrature
carrier modulation. Moreover, during the transmission of one symbol, the
constellation at the chip time presents no zero crossing, ensuring small envelope
variations for the transmitted modulated signal, of one given user. The
computation of the performances are related to the desired code "1" defined by:
This code "1" is then reduced to the complex version of the scrambling code.
VII. 1.2. Propagation channels
Models described by ETSI [20] in "Indoor A" (IA) and "Vehicular B" (VB)
environments are used. The channel impulse responses are constituted by 6 paths
with relative delays (to the first paths) and average power, summarized in the
following table:
IA (nsec)
50
110
170
290
310
(dB)
-3
-10
-18
-26
-32
VB (nsec) 0
0.3
8.9
12.9
17.1
20.0
-13
-10
-25
-16
(dB)
-2.5
Advances in
UMTS technology
117
The temporal spreading is very weak in IA, slightly superior to one chip time,
Tc; on the contrary, the temporal spread is important in VB, around 5 23, which
will permit equalizer performances in two very different situations. The
propagation delays are quasi-fixed (variation of % 3%). The complex
amplitudes a: of the paths are random, distributed as with a Rayleigh law for the
modulus and uniform law between 0 and 2n for the phases. We suppose the
complex amplitudes uncorrelated between two paths (index i) and from one
sensor to another (index l). This last hypothesis, more often admitted [lo], is
certainly a bit optimistic for the inter-sensor spacing looking at the mobile; we
will come back to this point in the performance analysis.
VII.1.3. Average performances (random channels)

We are going to evaluate, for one given model of channel, the performances of
the equalizers from theoretical formulas ( 2 4 , 2 9 , 3 0 , 3 2 )developed in Section V.
The frequency integrals are approximated by the rectangle method, with a
and atemporal oversampling of Tcl8 to form the delays

frequency step Af = 1
32Ts
of the paths. These performances take the form
(SINR)
1
(3)
- where PI
No
$1
represents the degradation on the SINR caused by interference for the Matched
Filter, Zero-Forcing and MMSE receiver. For the Zero-Forcing equalizer, PI also
represents the degradation of the BER.
We recall that these performance formulae suppose perfect knowledge of the
channel and perfect timing synchronization, they give then upper bounds which
will inevitably be degraded in real situations. Moreover, they are only valid for
a deterministic and time invariant impulse response, which does not correspond
to IA and VB models since they are random, mainly from the complex amplitudes
a:. Rather than calculating the performances for one specific observed sample
of the path amplitudes (supposed then deterministic, equal for example to their
quadratic average values), we are going to give average performances for the
whole possible occurence of impulse response. This classical process [17, 101
may be used to represent a practical situation with slowly time variant channel:
one occurence of af is associated with the channel impulse response at a given
instant. The channel variation is slow enough to be considered fixed during the
occurence and perfectly estimated. So, the interference term PI, as with the
energetic gain of the global waveform Y,,,~, become random variables. We are
now interested in the expected values of the SINR inverse (named INSR) and of
the Error Probability (for a),
computed from a very large sampling number:
where
0,
yll[ol =
Yll[OI
Eb
expresses the amplitude fading, the on-lining
representing the expected or average value. The degradation is due to two

different phenomena: the interference, through PI, and the amplitude fading
(spreading of the range of the probability density function of the normalized
-
energy B , through P,).

Eb
In a order to obtain the average degradation caused only by the interference,

a situation without fading is artificially introduced. In this situation, from one
occurrence to another, the waveform is still random, but its energetic gain yll,ol
is maintained at
- a constant value ql,,ol, which-fixes the receive average bit
energy to Eb = Eb and makes deterministic'3,(
Pf = 1.
In a situationwith fading, only the BER performances in Zero-Forcing are
plotted since the INSR curves are practically derived from those of the situation
superior or equal to 1. This
without fading, by an additional weighting factor
last observation is explained by equation (33) and by the weak correlation
between the fading term P, and the interference term P I , as shown in Figure 10
which plots the correlation coefficient related to the number of users, from the
is thus an indicator of the degradation due to the fading, even if it
model VB.
is well known that the average energetic indexes are not sufficient to express
quantitatively the Error Probability. Without interference (PI = l), with a singlepath channel subject to Rayleigh fading, and with uncorrelated envelopes from
&,
13,
one sensor to another, Pe asymptotically varies [ 171 with (l&
No
- L.
The slope of
the BER curve (with classical Logscales) is imposed by the diversity order, equal
to the number of sensors in this basic example.
Figure 10. Correlation coeBcient between p, and

VB channel model.
computed with the
With any random channel, we may still define an equivalent diversity order
Neq relative to a Rayleigh fading, as proposed in [9], from the variation
coefficient cv of the power received:
We then dispose of two indexes to measure the depth of the fading: cv and f
With respect to a Rayleigh fading with N independent diversity branches (with
same power), it can be shown that:
In order to facilitate the interpretation of the results, we indicate in the

following table the values (in decibel) of these two indexes for a number of
diversity orders:
Neq
cv (dB)
f(dB)
-1.5
-2.3
-3.5
-4.5
00
1.76
0.96
0.57
+ 00
+ 00
VII.2. Performance analysis

VII.2.1. Vehicular without fading
Figure 11 presents the results in VB with 1,2,3 sensors for a situation without
fading. Each elementary cell of the figure shows the results superimposed for 1,
2, 4, 8, 12 and 16 users in continuous line, from bottom to top. The dotted lines
give the theoretical ideal performances obtained with a single path channel
without fading.
Concerning the (INSR), we first observe a great improvement (diminution of
I) in MMSE or ZF (equivalent to MMSE for strongEb/N0)as compared with the single
matched filter. For the MF, the (INSR) curves reach a floor, for strongEb/N0and for
the large values of K, because of the interference. The improvement due to the
equalization is emphasized with 2 or 3 sensors, which brings it almost back to
ideal performances.
Concerning the BER in ZF, there is an important loss, around 15 dB at
Pe = 10-3 when the saturation of the number of users (K = 16) is approached.
The multisensor reception gives a new degree of freedom and reduces the
degradation to just 1 dB with 3 sensors.
Figure 11. VB performances without fading.
VII.2.2. Vehicular with fading

Figure 12 presents the results for a ZF receiver in a fading situation (top of the
figure). The upper dotted lines correspond to the performance with a single-path
channel subject to Rayleigh fading with only one receiving sensor, i.e. without
diversity. The importance of the fading process is embodied by the bottom plots
which indicate the scattering of the probability density of the channel energy,
related to its average value. We also have mentioned the indexes cv and f
Figure 12. VB performances with fading.
We observe with a single-sensor a degradation of 3 to 5 dB compared to the

curves of BER without fading the situation. The very important degradation
above 18 dB for 16 users is reduced to 3 dB with 3 sensors.
We notice, for one user and no spatial diversity, that the BER sensitivity is
better than in a single path fading situation. This is naturally explained by the
path diversity (resolved paths) of the Vehicular environment. Indeed we have for
a single-sensor cv = - 1.6 dB and, f, = 2 dB, which already corresponds to a
diversity order Neq of between 2 and 3. This order approaches 5 and 7 with 2 and
3 sensors.
VII.2.3. Indoor with fading
It is interesting to compare the fading situation previously noticed with those
observed in the IA environment (Figure 13) where there is almost no path
diversity and very few interferences in the MF outputs. With a single sensor, the
error probability is nearly equivalent to the one path fading situation for any
number of users, expressing weak I and strong f. The spatial diversity is then
Figure 13. IA performances with fading.
required, even if K is small, in order to decrease the range of the amplitude

fading: for one single sensor, cv - - 0.3 dB and f = 5 dB, which corresponds
to a diversity order Neq of very little, slightly above 1. This order approaches 2.5
and 4 respectively with 2 and 3 sensors.
This improvement would be less significant if the paths were correlated
from one sensor to another. Nevertheless, the Indoor environment, which
mostly needs spatial diversity, presents a very favourable situation in downlink
reception on the mobile [18, 9], due to the proximity between the scattering
objects and the receiver. In this case, one path can be modelled by a lot of
micro-paths arriving at the mobile sensor with quasily the same delay.
Depending on to the model developed in [18] for one macro-path, the
correlation between any field component envelope measured at two separate
points on the mobile is shown to be weak with a spatial separation above /2.
To the contrary, the reception situation on the base stationin "uplink" would be
very different because of a geometrical asymmetry and would require a
spacing above 50 . for obtaining weak correlations.
In conclusion, a coherent multi-element antenna in reception improves the

transmission by fighting against both frequency selectivity and amplitude
fading.
Note: In terms of link budget, the presented performances should benefit
from an additional improvement of l0log(L) dB due to the antenna gain. All the
curves were plotted in the function of the global energy of all the sensors.
VIII. Conclusion
This study has permitted, by taking a frequency interpretations approach, to
recall non-exhaustively the linear detection structures in the downlink of a TDCDMA system. We have formulated the expressions and performances of this
detector from the representation in the symbol-band, based on the spectral
matrix of the aliased cross-channels. Finally, we have applied the performances
expressions to the VB and IA environment models. These performances constitute
upper bounds for real situations because of the inaccuracy in channel estimation
and in synchronization operation.
From the VB model, frequency selective, we have first illustrated the benefit
of the joint detection as compared with the simple desired waveform matched
filter, especially when the code margin is little. We have also underlined the clear
performance improvement for a coherent reception with 2 or 3 sensors. In this
selective channel, the multi-sensor reception permits the decrease of interference
and to facilitation of the inversion performed by the equalizer.
From the IA model, weakly selective but also very poor in temporal diversity
(non-resolved paths), we notice a spectacular improvement of performances due
to the diversity gain of the multi-sensor, reducing the range of amplitude fading.
So, multi-sensor reception on the mobile is always recommended to fight
efficiently the two different types of distortion introduced by the channel:
frequency selectivity and fading process.
REFERENCES
[1] VAN ETTEN (W.), Maximum likelihood receiver for multiple channel transmission
systems. IEEE Transactions on Communications, pp. 276-283, (Feb. 1976).
[2] TURIN (G.L.), Introduction to spread-spectrum antimultipath techniques and their
application to urban digital radio, Proceedings of the IEEE, vol. 68, n 3,
pp. 328-353, (March 1980).
[3] PICKHOLTZ (R.L.), SCHILLING (D.L), MILSTEIN (L.B.), Theory of spread-spectrum
communications - A tutorial. IEEE Transactions on Communications, vol.com-30,
n 5, pp. 855-884, (May 1982).
[4] JUNG (P.), BLANZ (J.), Joint detection with coherent receiver antenna diversity in
CDMA mobile radio systems. IEEE Transactions on Vehicular Technology, vol. 44,
n 1, pp. 76-88, (Feb. 1995).
[5] MADHOW (U.), HONIG (M.L), MMSE interference suppression for direct sequence
spread-spectrum CDMA. IEEE Transactions on Communications, vol. 42, n 12,
pp. 3178-3188, (Dec. 1994).
[6] WANG (X.), POOR (H.V.), Blind multiuser detection: a subspace approach.
IEEE Transactions on Information Theory, vol. 44, n 2, pp. 677-690, (March 1998).
[7] BALABAN (P.), SALZ (J.), Optimum diversity combining and equalization in digital
data transmission with applications to cellular mobile radio-Part I: Theorical
considerations. IEEE Transactions on Communications, vol. 40, n 5, pp. 885-894,
(May 1992).
[8] BOURENNANE (S.), FAURE (B.), L.ACOUME (J.L.), Traitement d'antenne pour des sources
bande large. Annales des Tlcommunications, vol. 45, n 7-8, pp. 384-392,
(July 1990).
[9] DIOURIS (J.F.), ZEIDLER (J.), BULJORE (S.), Space-path diversity in CDMA using a compact
array. Annals of Telecommunications, vol. 53, n 11-12, pp. 425-434, (1998).
[10] BULJORE (S.), DIOURIS (J.F.), SAILLARD (J.), tude thorique d'un galiseur multicapteur associ au critre EQM pour le canal radiomobile (modle GSM),
GRETSI, Juan-les-Pins, pp. 557-560, (Sept. 1995).
[11] VILA (P.), PIPON (F.), PIREZ (D.), FETY (L), Filtrage adapt multidimensioonel pour
l'galisation d'un canal slectif en frequence et brouill. GRETSI, Juan-les-Pins,
pp. 553-556, (Sept. 1995).
[12] PICEONNAT (Y.), Alternative solutions for joint detection in TD/CDMA multiple access
scheme for UMTS. SPAWC Anapolis, pp. 329-332, (May 1999).
[13] Ros (L), JOURDAIN (G.), ARNDT (M.), Multi-sensor reception for TDD-UMTS mobile
terminal in indoor and pedestrian environments. 11th IEEE International
Symposium on PIMRC, London, pp. 1221-1225, (Sept. 2000).
[14] Ros (L.), Reception multi-capteur pour un terminal mobile TDD-UMTS. Report us
n8!2000, (May 2000).
Advances in
UMTS technology
125
[15] CAPELLANO(V.), Apport des traitements spatio-temporels pour les transmissions
numeriques en presence de trajets multiples. Application aux communications

acoustiques sous-marines, Phd thesis, wGrenoble, (May 1998).
(51, Multiuser detection. Cambridge University Press, (1998).

I161 VERDU
[17] PROAKIS(J.G.), Digital communications. McGraw-Hill, third edition, (1995).
[18] JAKES(W.C.), Microwave mobile communications. EEE
PRESS,
New-York, (1974).
(5.) et al: Les methodes a haute resolution, traitement d'antenne et

[19] MARCOS
analyse spectrale. Collection Traitement du Signal, Hermes, (1998).
[20] ETSI: chapter B.1.4.2, Channel impulse response model.
UMTS 30.03,
TR101-112
V3.2.0, (April 1998).

[21] 3GPP: Technical specification: spreading and modulation (TDD). TS25.223, V3.2.0,
(March 2000).
Chapter 5
Smart-antenna space-time
UMTS uplink processing for
system capacity enhancement
T. Neubauer
Technical University, Vienna, Austria
E. Bonek
Technical University and FTW, Vienna, Austria
I. Introduction
The European third generation mobile communication system is known as the
Universal Mobile Telecommunications System (UMTS). It will enable tomorrow's
wireless Information Society, to deliver high-value broadband information,
commercial and entertainment services to mobile users. UMTS will play a key
role in creating the future mass market for high-quality multi-media
communications that will approach 2 billion users worldwide by the year 2010
[1].
The main advantage of third generation mobile systems will be mixed
services with variable data rates. The customer profile of a 3G system will
consist of a large number of low-data-rate and voice users with the concurrent
service to a high-data-rate user population.
The frequency division duplex (FDD) mode of UMTS is based on a Wideband
Code Division Multiple Access (W-CDMA) technique. CDMA systems are
interference limited in nature and suffer severely from the near-far effects
because all users operate in the same frequency band simultaneously. This is
essentially more critical in the uplink of UMTS where all users are potential
interferers. The problems due to the near-far effect are reduced in the downlink
because of the better orthogonality of the individual signals [2].
Since higher bit-rate services experience lower processing gains, they have
to be sent with higher power levels in order to fulfil the quality requirements at
the receiver. This causes greater interference at the base station (BS) antenna and
an intensification of the near-far problem. Hence, the capacity will be reduced
due to the overall interference increase.
I.1.1 Rake receiver structures

Radio propagation in the land mobile channel is characterized by multiple
reflections, diffraction and attenuation of the signal energy. These are caused by
natural obstacles such as buildings, vehicles, hills, and the like, resulting in
multi-path propagation. The signal energy may arrive at the receiver over many
clearly distinguishable time instants. The delay profiles extend typically up to 1
or 2 sec in urban and suburban areas [4]. In the FDD mode of UMTS, the chip
duration is 0.26 sec. If the time difference of the multi-path components is at
least 0.26 sec, the W-CDMA receiver can separate these multi-path components
and combine them coherently (rake them up) to obtain multi-path diversity.
Rake receiver structures are well known from second generation CDMA
systems like IS 95 where they are implemented. Available literature on rake
receiver structures is vast and covers capacity evaluation as well as
implementation aspects [8, 12, 26, 28, 29, 30].
I.1.2 Smart-antennas
Smart- or adaptive-antennas have been recently applied to improve the capacity
and the performance of second generation wireless mobile communication
systems [17, 20, 21, 33]. In general, smart-antennas reduce the effects of
multipath fading and improve the signal-to-noise-plus-interference ratio (SNIR).
Additional advantages such as channel capacity increase in urban areas,
coverage extension in rural areas and spectrum efficiency increase in general can
be obtained with the aid of smart-antennas at the base station. Drawbacks are
cost, and both hardware and software complexity.
In this paper we investigate the effects/benefits of smart-antennas in the FDD
uplink by means of UMTS system-level simulations. A simplified model for
smart-antennas in a multi-service UMTS scenario is shown in Figure 1. We
assume that the cell serves several low data rate users and a few high data rate
users. The latter are indicated by mobile terminals with large screen and
keyboard.
The high data rate users heavily interfere with the far-off user equipment
(UE).1 With smart-antennas the interference at the BS can be reduced for this UE
by placing broad nulls in the direction of the high data rate users. Due to this,
both the coverage range (limited by the TX power of the desired UE) and the
received signal quality will increase.
Alternatively, smart-antennas will increase the received SNIR. Power control
algorithms designed to keep the reception quality at the same level will then
reduce the transmit power of the UES. This will lead to a overall decrease in the
multiple-access interference (MAI), which, in turn, leads to higher system capacity.
1. Note that there are some similar notations in the standard. MS (mobile station), ME (mobile
equipment) and UE (user equipment). They sound similar, but they have different meaning [35].
128 Smart-antenna space-time UMTS uplink processing
Figure 1. Smart-antenna patterns in a multi-service UMTS system with

high-data-rate interferers and desired low-data-rate users.
Taking all this into account, smart-antennas promise a clear advantage in
multi-service traffic scenarios in comparison with non-adaptive antenna
techniques. It is generally accepted that the adaptive antenna technique will be a
key enabler for the success of UMTS [34].
II. Data model

In the following we present the channel model used, signal model and the overall
system model.
II. 1. Channel model

An important part of the evaluation process of a system employing smart-antenna
arrays is the channel modeling. For system capacity investigations, flat fading
channel models are widely used [22, 23]. This is a gross simplification of the
nature of mobile radio propagation channels and does not capture the detailed
information of state of the art channel modelling [4]. Hence, we study the impact
of a channel model that includes both the spatial and the temporal distribution of
the energy reaching the receiver. Moreover we compare this with the results
obtained under flat fading assumptions.
II.I.I. Flat fading channel
The flat fading assumption implies that the delay spread introduced by the
channel is much smaller than the time duration of the transmitted symbols. In
our idealized flat fading channel the delay spread is zero and hence all incoming
waves are received with delay = 0.
It is well accepted that the received signal at the BS originates from scatterers
within a disk of radius r centred R metres from the BS around the UE [24].
Measurements in various urban environments have shown that the spatial
behaviour of the incoming signals at the BS can be modelled by a main direction
of arrival (DOA) in the azimuth and a corresponding angular spread [5].
The path loss is modelled based on the proposed models for UMTS system
simulations presented in [2] for pre-defined environments. Path loss correlation
between two up-link connections from a single UE to different BSS is expressed
by a correlation coefficient of 0.5 [10]. The log-normal large-scale fading is
2
modeled with a mean of In = 0 and a standard deviation of ln = 6 dB.
II. 1.2. Spatio-temporal channel model
Within the European research initiative COST 259 [4] several channel models
have been developed. They are aimed at UMTS and HIPERLAN, with particular
emphasis on adaptive antennas and directional channels. They have been
introduced in the third generation standardization process by 3GPP [3], and they
are used here in the following way.3
The path loss again is modelled based on the proposed models for UMTS
system simulations presented in [2] for pre-defined environments. Path loss
correlation and the log-normal large-scale fading are modelled in the same way
as in the flat fading case.
In order to test advanced antenna array systems, the main focus of the
channel model has to be on the azimuthal and temporal properties. The statistics
of the delay and azimuth of the impinging waves as well as their expected power
are especially of great interest. According to [4] and [5] we define a Laplacian
shape for the Azimuth Power Spectrum (APS) seen at the BS antenna. For the
Delay Power Spectrum (DPS) we assume the widely used one-sided exponential
decaying function [5]. Both can be written as
where , , t and denote the azimuth towards the UE seen at the BS, the azimuth
spread, the delay and the delay spread, respectively. For typical urban
environments the azimuth power delay spectrum can be written as
2. Enlarging the standard deviation to ln = +10 dB has negligible impact on the results.
3. Within 3GPP [3] only thee temporal domain is considered till now.
Figure 2. Assumed APS and DPS (according to the 3GPP-COST259 proposal).

= 0.5 sec, O = 0 sec and = 8o are assumed.
In our model we assume a tapped delay line model for a typical urban
scenario. We consider a single scattering cluster and L impinging multipathcomponents. The delay spread is = 0.5 sec and the start of the cluster is at
0 = 0 sec [3]. For the azimuth domain we assume a spread of a, = 8 around
0, where 0 corresponds to the azimuth toward the UE. In Figure 2 both APS a
DPS are shown schematically and for the parameters above.
II.2. System model

Following the results of [9], 19 sites for single cell investigations have to be
considered in the system model. For each site we assume 3 sectored cells.
Hence, a total of 57 cells altogether has been taken into account.
An example for a single element pattern, which approximates a commercial
UMTS antenna pattern, is shown in Figure 3. We assume that it is identical for all
antenna elements.
The main system and simulation parameters are summarized in Table I.
Figure 3. Single-element radiation pattern.

Table I. Main simulation parameters.
Number of cells
57
Inter-BS distance
1000m
Background noise floor

Path loss model
Log-normal large-scale fading
Path loss correlation
- 105 dBm [36]

Macro cell [2]
1n = 0,1n = 6dB
0.5
Max. UE TX power
21 dBm
TPC dynamic range
80 dB
= 0, p= 0.5dB
Power control error

No. of antenna elements
Inter-element spacing
Number of impinging rays
Delay spread
Azimuth spread
1 and 4
d = /2
10
= 0.5sec
= 8
II.3. Signal Model

Following the notation of Bello [6], the azimuth-delay spread function of the
channel for each path l from UE k to BS is given by
where V Pl,k,, and 9l,k, denote the amplitude and the phase of the l-th received
signal path at a single antenna element impinging with delay l,k, and azimuth
l,k,.(,) is the Dirac delta function. The array response vector for a single
UE k at BS , for the delay is thus given as
where cl,k,, is the array steering vector of a wave impinging from an azimuth
direction l,k, at an M-element uniform linear array with an inter-element
spacing d.
In the FDD mode of UMTS, the time a wavefront takes to pass through the array
is much smaller than the chip interval Tc. Therefore, the narrowband assumption
for antenna arrays is valid. This makes it possible to model the time delays as
phase shifts. The array steering vector can thus be written as
where is the wavelength of the impinging signal. If we assume that we know

perfectly the channel, the spatial covariance matrix of the k-th user in the cell ,
at a certain delay becomes:
In the same way we define the spatial interference and noise covariance
where 2N and I denote the noise variance and M x M the identity matrix,
respectively.
In the case of the flat fading channel model, we include an angular spread.
Borrowing the technique from [25], we derive the spatial covariance matrix of
the k-th user at a certain instant in time as
with
where O and [.]pq denote the Schur-Hadamard element-by-element matrix

product and the pq-th element of a matrix, respectively. k, indicates the
angular spread of the corresponding DOA k,. Applying (9) to the interference
covariance matrix we get
III. Space-time processing

Space-time processing in third generation systems can be seen as an evolution
of more "traditional" receiver structures. Space-time processing combines timeonly and space-only processing. Time-only processing corresponds to equalizers
that use a weighted sum of signal samples (RAKE receiver), and space-only
processing corresponds to simple beamforming that uses a weighted sum of
antenna outputs. Examples for available literature on space-time processing are
[12, 13, 19,27,28,31].
A block diagram of our space-time BS receiver structure is shown in
Figure 4. In the literature, this receiver structure is also named the "decoupled
space-time rake", the "beamformer rake", the "2D-rake" and the "vector Rake single beamformer".
III.l. Spatial processing

For the implementation of combining algorithms of the multiple antenna signals,
it is important to know whether the spatial interference and noise covariance
matrix is diagonal or not [14, 11], meaning the interference is white or not. In the
case of a large number of users in the system, spatial whiteness can be generally
expected [7, 8]. The UMTS radio network controller (RNC) will reduce the
transmit power level of each individual UE to a minimum, even in the case of a
few high-data-rate users. Hence, singular interferers with high power levels,
which would be represented by high off-diagonal correlation, do not occur.
Figure 4. BS space-time receiver structure.
We investigated whether interferer whiteness is fulfilled or not. An example

of a long-term interference and noise covariance matrix with only few high-datarate users is given in Figure 5. We observe that the covariance matrix has a
diagonal shape indeed. But we also see that the spatial covariance matrix is not
stringently diagonally dominant [32] and hence spatial whiteness is not given
per se in a multi-service environment. This corroborates the intuition that single
"very" high data users contradict the assumption of a spatially white covariance
matrix.
The general problem in spatial processing is to find a weight vector for each
user k in order to maximize the signal-to-noise-plus-interference ratio SNIR k in
the uplink.
Without making any assumptions about the spatial covariance matrix behaviour,
the solution for this problem is to take the generalized eigenvector
corresponding to the largest eigenvalue [R(s)k, Q(s)k,] [15]. This well-known
beamforming concept corresponds to maximizing the signal-to-noise-plusinterference ratio SNIR in the uplink [14] and the estimated ratio of signal to
interference power for the k-th UE. From
Figure 5. Long-term interference and noise covariance matrix seen

at the antenna elements of the central BS.
we obtain the optimum spatial weights for the k-th UE in cell as a function of
the delay . Since R contains all signal information, the received signal power
after spatial processing for the n-th delay tap is given by
III.2. Temporal processing

Applying optimal spatial weights to the receiver structure shown in Figure 4, we
obtain optimum SNIRS for each delay tap at the rake receiver input for each
individual user.
For the temporal consideration we assume a wide-sense-stationary (wss)
channel with uncorrelated scattering (us) only. The complete short-term channel
estimate in time can be obtained by correlation with the pilot sequence, which is
part of the control channel in the UMTS FDD mode. The consequence of the us
condition is that the contributions of the temporal taps selected for further
processing steps are uncorrelated.4
An example for a temporal signal covariance matrix is given in Figure 6.
Since we only have a limited number N, of rake fingers, the taps with the
highest signal power are selected. As mentioned above, we can assume temporal
whiteness of the interfering signals. Thus, the optimum output of the rake
receiver can be achieved by maximum ratio combining (MRC) of the selected
fingers [14]. The weights for the rake fingers are thus given by
where hn(T) and ()* denote the temporal channel impulse response and complex
conjugate, respectively.
IV. Simulations
For investigations of system behaviour, we now have to consider a number of
cells. As already mentioned, we consider 19 3-sectored sites, leading to a total
of 57 cells. In doing so, we correctly estimate the inter-cell interference at the
inner site and thus we are able to come up with reliable conclusions at the end
[9].
Taking the whole network into account, we calculate the interference seen at
the BS antenna for each user. Since we now have a total number of K users in the
entire network, we easily find from (8)
4. Two adjacent taps are usually not completely uncorrelated due to filtering and oversampling.
Analysis in [16, chapter 29.3 "The selective rake receiver"] has shown that the correlation is in the
range of 0.02 for raised cosine pulse shapes. Thus, we assume two adjacent taps to be uncorrelated.
Figure 6. Long-term temporal signal covariance matrix.
the spatial covariance matrix including the total network interference for the
k-th user belonging to cell . Following the description above and repeating the
procedure for = 1, ..., Z (Z = 57) cells, we can finally calculate the quality
requirement at the BS antenna connector for each user in the network.
Introducing the processing gain GP as a function of the service of the k-th
user
with the modulation bandwidth W and the user data rate Rk, , we obtain the ratio
of combined received energy per information bit to the effective noise power
spectral density, Neff, at a single point in time.
RSCP k,
and
ISCP k ,
are the Received Signal Code Power and the
Interference Signal Code Power (including noise) of the k-th user in cell after
beam forming and rake combining.
For the computer simulation of the above model, we use a stationary MonteCarlo approach [9, 22, 23]. The schematic flowchart of the system simulator is
shown in Figure 7.
The main simulation parameters are given in Table I.
V. Results and discussion

This section is structured as follows. First we give the definition of the system
capacity bound and the system load in the uplink of a CDMA system. Then we
discuss the results of space-only processing and finally of space-time processing.
V.I.I. System load definition
The maximum load, i.e., the maximum capacity of a single cell in the UMTS FDD
uplink, is defined as the number of users corresponding to a certain rise in
multiple-access interference MAI over the thermal background noise [2]. This
"noise rise" is the increase in the uplink noise at the BS antenna connector due to
the uplink interference. It is expressed as a dB figure relative to the thermal noise
floor (more precisely, the thermal noise including the man-made noise). The 95%
confidence value of this noise rise is taken over all Monte Carlo runs and all
monitored cells. Our chosen design target value for this figure will be 6 dB. This
is equivalent to 75% of the pole-capacity loading of a CDMA system [2]. The pole
capacity is the maximum theoretical capacity of an isolated CDMA cell [26].
Figure 7.
Simplified flowchart of the
system simulator.
Figure 8 shows the relationship between uplink noise rise seen at the BS antenna
and the system load in %. The term is defined as
where SIRloaded and SIR empty are the signal-to-interference-ratio at the BS antenna
in a loaded cell (or system) and in an empty cell (or system), respectively [22].
V.1.2. Space-only results
We consider the flat fading propagation channel as described in Section 2. The
individual multi-path components receive at = 0 at the BS antenna.
We now make a comparative performance evaluation of simplified multiservice scenarios of the UMTS FDD uplink. We consider two systems, one equipped
with single-element antennas and the other with four-element antenna arrays at
all base stations within the network. Further we assume a service mix of speech
users with a net data rate of 12.2 kbit/sec and 64 kbit/sec data users. As
schematically explained in Figure 1, we expect that the gain of smart-antennas
over the single-element antenna is higher, if the percentage of high-data-rate
users in the network is high. Hence, we compare the noise rise of both systems
for various service mixes, starting with speech users only.
Figure 8. Theoretical interrelationship between the noise rise

at the BS antenna and the system load in %.
We can observe from Figure 9 that the shape of the noise rise in case of
speech users only is a smooth one for both systems. Taking a closer look, e.g.,
at the noise rise at the 50 users/site-level, we find a value of 5.8 dB for the nonadaptive-antenna system, while the noise rise for the smart-antenna network is
only about 2.8 dB. This 3 dB gain in the noise rise is in the range of what was
expected for CDMA systems with speech users only [18, 19].
The noise rise is increasing when we add more users to the single-antennaelement system. In contrast, the noise in the smart-antenna network rises only
slightly as a function of both the number of speech users and the increased
number of 64 kbit/sec-data users.
CDMA systems are usually planned to operate at a system load of = 50%
(as defined in Equation 19, [22]). For this operation point, Figure 10 defines the
plane of possible business cases, for both systems. A business case is the number
of users for the different services in the system area. The dashed lines (with the
marker for the single antenna case, without marker for the smart-antenna case)
indicate the respective capacity maximum, i.e. for 50% of system loading. To the
left of these lines, we can look for viable business cases. We can obtain an
ensemble of possible user ratios in our system. One such user ratio (a "business
case") with 54% data users will serve a total of 15 users (8 64 kbit/sec data, and
7 12.2 kbit/sec speech) in the conventional system, and about 48 users (26-64
kbit/sec data and 22 12.2 kbit/sec speech) by applying smart-antennas to the Bss.
The actual capacity gain, of course, depends on individual realization of the
network.
This directly leads to the question of the effective throughput as a function of
the system load and the percentage of high data rate users in the system.
Figure 9. Noise rise
as a function of both
the service mix of
speech (12.2 kbit/sec)
and 64 kbit/sec data
users and the number
of users per site.
Two curves are
shown. One for the
noise rise for the
system using
4-element antenna
arrays, and a second
one for the single
element antenna.
The latter completely
overlays the first one.
Figure 10.
Boundaries for
a single-element
antenna system
(lower left area) and
for a smart-antenna
system (upper left area)
as a function of the
system load.
The boundary is shown
for a system load of
= 50%. Arrows
indicate a specific
business case
(54% data users).
As indicated in Figure 11, the maximum effective throughput for individual
business case realizations depends on the system load as well as on the service
mix. For smart-antenna systems, the maximum throughput is steadily increasing
with the number of high-data-rate users, but that is not the case for conventional
antenna systems. This fact will be a main challenge for future UMTS operators
and network planners. They have to optimize the network depending on the
requirements of an increasing number of future data users, as well as on the
system performance.
The plateaus, the kinks and bumps, are a consequence of our taking the 95%
percentile value of CDFS.
Figure 11. Effective
throughput vs. the percentage of high-data-rate users

in the system area.
V.1.3. Space-time results

For the investigations on space-time performance on system-level we take the
spatio-temporal channel model proposed in Section 2. Moreover we assume
ideal channel estimation for the weight calculations.
We investigated a service mix of 40% speech users and 60% 64 kbit/sec
data users. Figure 12 shows the results, comparing a single antenna and the
4-element antenna array case in a flat fading channel and the COST259
environment. For the temporal combiner we assumed a maximum of four
selective rake fingers. Our investigation reveals a diversity gain due to rake
finger assignment. Thus, the noise rise in the COST259 channel is consistently
smaller than in the flat fading environment.
We also investigated 1, 4 and 6 antenna element uniform linear arrays with
2 and 4 rake fingers for the temporal combining as schematically shown in
Figure 4. The results for a service mix of 40% speech users and 60% 64 kbit/sec
data users are presented in Figure 13. We assumed a delay spread of 0.5 sec
[3] (about 2TC ) and a number of 10 impinging multi-path components.
In the case of 2 rake fingers only, the results are worse than in the flat fading
environment, which corroborates the intuition that two rake fingers sum up only
a portion of the power received [16, 29].
In general, the performance of rake receivers and the number of necessary
rake fingers depend on the environment. In narrowband channels a low number
of rake fingers is sufficient to sum up the signal energy. If we consider a urban
channel with a high delay spread, more rake fingers are required [29]. This can
be observed from Figure 13, where four rake fingers perform better than two in
any antenna case.
Figure 12. Noise rise vs. the number of users per cell for aflat fading and
the proposed spatio-temporal channel model.
Figure 13. Noise rise in the central cells. The BS are equipped with
1,4 or 6 antenna elements and the temporal rake receiver combines 2 or 4 taps.
Future UMTS base stations will provide at least four rake fingers per receiver
branch. Thus, the flat fading assumption turns out to be a lower bound for the
capacity estimation for both conventional and smart-antenna systems,
respectively.
VI. Conclusions
The main focus of this study is the benefit of smart-antennas in the uplink of a
multi-service UMTS FDD system. We demonstrated the importance of arealistic
channel model accounting for spatio-temporal characteristics of mobile radio
channels for a fair assessment of smart-antenna benefits.
We applied the capability of antenna array processing to our stationary
Monte-Carlo UMTS system simulation approach. Based on this we investigated
the mixed service capacity in the uplink. We compared and evaluated the
performance of space-only and space-time processing under both flat fading and
spatio-temporal channel conditions. We investigated systems with singleelement antennas, four- and six-element antenna arrays at all base stations within
the network. Also we looked at the effect of a limited number of rake fingers in
an urban environment. Building on the results in Section 5 we conclude:
There can be a substantial increase in system capacity by incorporating
antenna arrays at the BS.
The higher the number of high data rate users is, the more will smartantennas outperform conventional antenna systems.
The maximum effective throughput for individual business case

realizations depends on the service mix. For smart-antenna systems the
maximum throughput rises with the number of high-data-rate users in the
system. This is not the case for single antenna networks.
Contrary to GSM, the maximum throughput is a function of the system load.
Therefore, network planning and optimization will be a main challenge for
future UMTS operators.
The flat fading assumption is a lower bound for the capacity estimation in
urban scenarios.
Overall, we have shown a significant increase of mixed service capacity of
UMTS uplink by smart-antennas and thus we strongly recommend their use at the
BS.
Acknowledgement
The authors would like to thank Klaus Hugl, Werner Weichselberger and Thomas
Baumgartner for fruitful discussions. They also acknowledge the useful editorial
comments by Martin Tltsch and the reviewers of this paper.
REFERENCES
[1] UMTS-Forum, "The Future Mobile Market", Report n 8, www.umts-forum.org,
(March 1999).
[2] 3gpp, "RF System Scenarios -TS 25.942", http://www.3gpp.org, (April 2000).
[3] 3GPP, "Deployment aspects - TS 25.943", http://www.3gpp.org, (April 2000).
[4] CORREIA (L.H.), ed., Wireless Flexible Personalized Communications - COST 259:
European Co-operation in Mobile Radio Research. J. Wiley & Sons, (March 2001).
[5] PEDERSEN (K.I.), MOGENSEN (P.E.), FLEURY (B.H.)," A stochastic model of the temporal
and azimuthal dispersion seen at the base station in outdoor propagation
environments", IEEE Transactions on Vehicular Technology, 49, n 2, pp. 437-447,
(March 2000).
[6] BELLO (P.A.), "Characterization of Randomly Time-Variant Linear Channels",
IEEE Trans. on Communication Systems, CS-11, pp. 360-393, (Dec. 1963).
[7] DELL'ANNA (M.), AGHVAMI (A.H.), "Performance of optimum and suboptimum
combining at the antenna array of a W-CDMA system", IEEE Journal on Selected
Areas in Communications, 17 n 12, pp. 2123 -2137, (Dec. 1999).
[8] NAGUIB (A.F.), PAULRAJ (A.), "Performance of wireless CDMA with M-ary orthogonal
modulation and cell site antenna arrays", IEEE Journal on Selected Areas in
Communications, 14, n 9, pp. 1770 -1783, (Dec. 1996).
[9] NEUBAUER (T.), BAUMGARTNER (T.), BONEK (E.), "Necessary and Sufficient Network Size
for Pole Capacity Estimation in UMTS FDD", ECWT 2000, Paris, France, pp. 131-134,
(Oct. 2000).
[10] VITERBI (A.J.), VITERBI (A.M.) "Other-Cell Interference in Cellular Power-Controlled
CDMA", IEEE Transactions on Communications, 42, n 2/3/4, pp. 1501-1504,
(Feb.
1994).
[11] KOPSA (K.), WEINMANN (R.), BRAUN (V.), TANGEMANN (M.), "Performance of Multiple
Antenna Combining Algorithms in the Uplink of UMTS/FDD", Globecom 2000-11-27.
[12] A. PAULRAJ (A.), PAPADIAS (C.B.), "Space-time processing for wireless
communications", IEEE Signal Processing Mag., 14, pp. 49-83, (Nov. 1997).
[13] THOMSON (J.), GRANT (P.), MULGREW (B.), "Smart antenna arrays for CDMA systems",
IEEE Personal Commun. Mag., pp. 16-25, (Oct. 1996).
[14] WINTERS (J.H.), "Optimum Combining in Digital Mobile Radio with Co-channel
Interference", IEEE Journal on Selected Areas in Comm., SAC-2, n 4,
pp. 528-539, (1984).
[15] GERLACH (D.), PAULRAJ (A.), "Adaptive transmitting antenna methods for
multipath environments ", IEEE Global Telecommunications Conference,
GLOBECOM '94, 1, pp. 425-429, (1994).
[16] MOLISCH (A.F.), "Wideband Wireless Digital Communications", Prentice-Hall,
(2000).
[17] KUCHAR (A.), TAFERNER (M.), TANGEMANN (M.), HOEK (C.) "Field trial with a
GSM/DCS1800 smart antenna base station" Vehicular Technology Conference,
Amsterdam, Holland, 1, pp. 42-46, (1999).
[18] LIBERTI (J.C.), RAPPAPORT (T.S.), "Analytical Results for Capacity Improvements in
CDMA", IEEE Trans. on VT, 43, n 3, pp. 680-690, (Aug. 1994).
[19] NAGUIB (A.F.), PAULRAJ (A.), KAILATH (T.), "Capacity Improvement with
Base-Station Antenna Arrays in Cellular CDMA", IEEE Trans. on VT, 43,
pp. 691-698, (Aug. 1994).
[20] HAGERMAN (B.), MAZUR (S.), "Adaptive Antennas in is-136 Systems", IEEE 48th
Vehicular Technology Conference 1998, Ottawa, Canada, pp. 2282-2286,
(May 1998).
[21] BORMANN (R.), FRERICH (M.), AHRENS (F.), HENB (T.), DAM (H.), BERG (M.), ANDERSSON (S.),
ALMQVIST (D.), "Functional Tests of adaptive antenna base stations for GSM",
3rd EPMCC 1999, Paris, France, pp. 323-327, (March 1999).
[22] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), JSBERG (M.), "Static simulator for
studying wcdma radio network planning issues", IEEE 49th Vehicular Technology
Conference 1999, VTC' 99 spring, Houston, USA, (May 1999).
[23] DEHGHAN (S.), LISTER (D.), OWEN (R.), JONES (P.), "W-CDMA capacity and planning
issues", IEE Electronics & Communication Engineering Journal, pp. 101-118,
(June 2000).
[24] FUHL (J.), MOLISCH (A.F.), BONEK (E.), "Unified channel model for mobile radio
systems with smart antennas", IEE Proceedings - Radar, Sonar and Navigation,
145, n 1, pp. 32- 41, (Feb. 1998).
[25] RIBA (J.), GOLDBERG (J.), VAZQUEZ (G.), "Robust Beamforming for Interference
Rejection in Mobile Communications", IEEE Trans. on Signal Processing, 45, n 1,
pp. 271-275, (Jan. 1997).
[26] LEE (J.S.), MILLER (L.E.), "CDMA Systems Engineering Handbook", Artech House Inc.,
(1998).
[27] BRUNNER (C), HAARDT (M.), NOSSEK (J.A.), "On space-time RAKE receiver structures
for WCDMA", Conference Record of the Thirty-Third Asilomar Conference on
Signals, Systems, and Computers, 2, pp. 1546-1551, (1999).
[28] KHALAJ (B.H.), PAULRAJ (A.), KAILATH (T.), "2D RAKE receivers for CDMA cellular
systems", IEEE Global Telecommunications Conference, GLOBECOM '94, 1,
pp. 400-404, (1994).
[29] WIN (M.Z.), CHRISIKOS (G.), MOLISCH (A.F.), SOLLENBERGER (N.R.), "Selective Rake Diversity
in Multipath Fading with Arbitrary Power Delay Profile", IEEE Global
Telecommunications Conference, GLOBECOM 2000, San Francisco, (27 Nov.-1 Dec.
2000).
[30] PROAKIS (J.G.), "Digital Communications", McGraw Hill Book Comp. Inc.,
3rd edition. New York, (1995).
[31] PEDERSEN (K.I.), "Antenna Arrays in Mobile Communications", Ph.D. Thesis,
Center for PersonKommunikation, Aalborg University, Denmark, (Jan. 2000).
[32] LANCASTER (P.), TISMENETSKY (M.), "The Theory of Matrices", Second Edition,
Academic Press, (1985).
[33] CUPO (R.L.), GOLDEN (G.D.), MARTIN (C.C.), SHERMAN (K.L.), SOLLENBERGER (N.R.),
WINTERS (J.H.), WOLNIANSKY (P.W.), "A four-element adaptive antenna array for
is-136 PCS base stations" IEEE 47th Vehicular Technology Conference, 3, (1997).
[34] ADACHI (F.), SAWAHASHI (M.), SUDA (H.), "W-CDMA: Performance evaluation and
future enhancement", 3rd European Personal Mobile Communications
Conference, EPMCC, Paris, 9-11, pp. 429-434, (March 1999).
[35] 3Gpp, "Vocabulary for 3gpp Specifications - TS 21.905", http://www.3gpp.org,
(April 2000).
[36] NEUBAUER (T.), JGER (H.), FUHL (J.), BONEK (E.), "Measurement of the Background
Noise Floor in the UMTS FDD Uplink Band", 4th European Personal Mobile
Communications Conference, EPMCC, Vienna, (Feb. 2001).
Chapter 6
Radio network planning process

and methods for W-CDMA
J. Laiho and A. Wacker
Nokia Group, Helsinki, Finland
I. Introduction
As the launch of third generation technology approaches, operators are forming
strategies for the deployment of their networks. These strategies must be
supported by realistic business plans both in terms of future service demand
estimates and the requirement for investment in network infrastructure.
Evaluating the requirement for network infrastructure can be achieved using
system dimensioning tools capable of assessing both the radio access and the
core network components. Having found an attractive business opportunity,
system deployment must be preceded by careful network planning. The network
planning tool must be capable of accurately modelling the system behaviour
when loaded with the expected traffic profile. The third generation cellular
systems will offer services well beyond the capabilities of today's networks. The
traffic profile, as well as the radio access technology itself, form the two most
significant challenges when dimensioning and planning a W-CDMA based third
generation system. The traffic profile describes the mixture of services being
used by the population of users. There are also specific system functionalities
which must be modelled including fast power control and soft handover. In order
to accurately predict the radio coverage the system features associated with
W-CDMA must be taken into account in the network modelling process. Especially
the channel characterization, and interference control mechanisms in the case of
any CDMA system must be considered. In W-CDMA network multiple services coexist. Different services (voice, data) have different processing gains, Eb/No
performance and thus different receiver SNR requirements. In addition to these
the W-CDMA coverage depends on the load characterization, hand over
parameterization, and power control effects. In current second generation
systems' coverage planning processes the base station sensitivity is constant and
the coverage threshold is the same for each base station. In the case of W-CDMA
the coverage threshold is dependent on the number of users and used bit rates in
all cells, thus it is cell and service specific.
The W-CDMA planning process can be divided into three phases: initial
planning (dimensioning), detailed radio network planning and network
operation and optimization. Each of these phases requires additional support
functions like propagation measurements, Key Performance Indicator
definitions etc. In a cellular system where all the air interface connections
operate on the same carrier the number of simultaneous users is directly
influential on the receivers' noise floors. Therefore, in the case of UMTS the
planning phases cannot be separated into coverage and capacity planning. In the
case of the post-second generation systems data services start to play an
important role. The variety of services requires the whole planning process to
overcome a set of modifications. One of the modifications is related to the
quality of service (QoS) requirements. So far it has been adequate to specify the
speech coverage and blocking probability only. Also one has to consider more
and more the indoor and in-car coverage probabilities. In the case of UMTS the
problem is slightly more multi-dimensional. For each service the QoS targets
have to be set and naturally also met. In practice this means that the tightest
requirement shall determine the site density. In addition to the coverage
probability the packet data QoS criteria are related to acceptable delays and
throughput. Estimation of the delays in the planning phase requires good
knowledge of the user behaviour and understanding of the functions of the
packet scheduler.
Common features between second and third generation coverage prediction
also exist. In all the systems both of the links have to be analysed. In current
systems the links tend to be in balance whereas in the case of third generation one
of the links can be higher loaded than the other, and thus either one of the links
could be limiting the cell capacity or coverage. The propagation calculation is
basically the same for all standards, with the exception that different propagation
models could be used. Another common feature is the interference analysis. In the
case of W-CDMA this is needed for the loading and sensitivity analysis, in the case
of TDMA/FDMA it is essential for frequency allocation. In order to fully utilize the
W-CDMA capabilities, a thorough understanding of the W-CDMA air interface is
needed from the physical layer to the network modelling, planning and
performance optimization.
In this paper the pre-operational phase of the W-CDMA planning process, as
depicted in Figure 1 in detail, is discussed. Section II concentrates on the initial
planning issues. The W-CDMA link budget is introduced and it is demonstrated
how different services and their QoS requirements impact on the site density
estimate. In Section III a static radio network simulator is introduced. The
148 Radio network planning process
Figure 1.
W-CDMA radio
network planning process.
methodology required in the coverage and capacity estimation for W-CDMA is

described for both uplink and downlink. The aim of Section IV is to demonstrate
the accuracy of dimensioning: an example area is dimensioned and the average
site distance is determined. The dimensioning result is compared with the
outcome of a static network simulation. In Section V the analysis methods of the
static simulator are verified. The verification was performed with a dynamic
system simulator. The paper is concluded in Section VI.
11. Initial planning, system dimensioning

Initial planning (i.e. system dimensioning) provides the first and most rapid
evaluation of the network element count as well as the associated capacity of
those elements. This includes both the radio access network as well as the core
network. This paper focuses upon the radio access part solely. The target of the
initial planning phase is to estimate the required site density and site
configurations for the area of interest. Initial planning activities include radio link
budget (RLB)and coverage analysis, capacity estimation, and finally, estimation
for the amount of base station hardware and sites, radio network controllers
(RNC),equipment at different interfaces, and core network elements. The service
distribution, traffic density, traffic growth estimates and QoS requirements are
essential already in the initial planning phase. In the initial planning phase the
quality is taken into account in terms of blocking and coverage probability. RLB
calculation is done for each service, and the tightest requirement determines the
maximum allowed isotropic path loss.
11.1
W-CDMA
specific items in the radio link budget
In this section the W-CDMA uplink and downlink budgets are discussed. To
estimate the maximum range of a cell a RLB calculation is needed. In the RLB the
antenna gains, cable losses, diversity gains, fading margins, diversity gains etc.
are taken into account. The output of the RLB calculation is the maximum
allowed propagation path loss which in return determines the cell range and thus
the number of sites needed. There are a few W-CDMA specific items in the link
budget if compared with the current TDMA based radio access system like GSM.
These include interference degradation margin, fast fading margin, transmit
power increase and soft handover gain.
The interference degradation margin is a function of the cell loading. The
more loading is allowed in the system, the larger interference margin is needed
in uplink, and the smaller is the coverage area. The uplink loading can be derived
as follows, for simplicity the derivation is performed with service activity v = 1.
To find out the required uplink transmitted and received signal power for a
mobile station MSk connected to a particular base station BSn, the basic CDMA
Eb/No equation is used. The usual, slightly theoretical, assumption is that I oth, the
interference received from the MSs connected to the other cells, is directly
proportional (proportionality constant i) to Iown, the interference received from
the MSs connected to the same BSn as the desired MS. Assume that the MSk uses
bit rate Rk, its Eb/No requirement is k and the CDMA modulation bandwidth is W.
Then the received power of the k-th mobile, pk, at the base station it is connected
to, must be at least such that
where Kn is the number of MSs connected to BSn, N = NoW = NfkToW is the

noise power in the case of an empty cell, Nf is the receiver noise figure, K is the
Boltzmann constant and To is the absolute temperature.
The inequalities in (1) are slightly optimistic because it is assumed that there
is no interference from the own signal. In reality this is not exactly true in multipath propagation conditions. Equation (1) is however still chosen to avoid taking
multi-path interference into account twice, i.e., the Eb/No requirements
determined from link level simulations are presented so that No means only noise
and multi-path interference is visible in higher Eb/No requirement to a certain
BER performance. Solving the inequalities as equalities means solving for the
minimum required received power (sensitivity), pk:
If the equations in (2) are summed over the mobile stations connected to BSn
then
since Iown = pk . If loading is defined as
this loading definition can be enhanced to include sectorization gain, ,, and

service activity, v:
(5)
In [12] the uplink loading is estimated using equation
Advances in
UMTS technology
151
where m is the number of services used. The difference between equations (5)
and (6) are due to the fact that (6) does not include sectorization gain and that in
the derivation starting from equation (1) the denominator is I,,,, - p k + do,,, +N
rather than I,,, + il,,,,,, +N, which is only the case when p k << lo,.
The downlink dimensioning is following the same logic as the uplink. For a
selected cell range the total base station transmit power ought to be estimated.
In this estimation the soft handover connections must be included. If the power
is exceeded either the cell range ought to be limited, or number of users in a cell
has to be reduced. For downlink the loading (T&
is estimated based on
where Lpmi is the link loss from the serving BS m to MS i, Lpni is the link
loss from another B S n, to M S i , pi is the transmit E,/N, requirement for the
MS i, including the SHO combining gain and the average power raise caused
by fast power control, N is the number of base stations, I is the number of
connections in a sector and ai is the orthogonality factor depending on
multi-path conditions (a= 1: fully orthogonal).
The term
Lpmi
defines the iDL.
Direct output of the downlink RLB is the single link power required by a user
at the cell edge. The total base station power estimation
must take into account
multiple communication links with average (ZPmi)
distance from the serving
base station. Furthermore, the multi-cell environment with orthogonalities ai
should be included in the modelling. More on the downlink loading and transmit
power estimations can be found in [13]. In the RLB calculation in uplink direction
the limiting factor is the mobile station transmit power, in downlink direction the
limit is the total base station transmit power. When balancing the uplink and
downlink service areas both links must be considered.
The interference degradation margin to be taken into account in the link
budget due to a certain loading q (either in uplink or downlink) is
Fast fading margin or power control headroom is another CDMA specific item
in the RLB. Some margin is needed in the mobile station transmission power for
maintaining adequate closed loop fast power control in unfortunate propagation

conditions like the cell edge. This is applicable especially for pedestrian users
where the Eb/N0 to be maintained is more sensitive to the closed loop power
control. The power control headroom has been studied more in [6] and [7].
Another impact of the fast power control is the transmit power increase. In the
case of a slowly moving mobile station the power control is able to follow the
fading channel and the average transmitted power increases. In its own cell this
is needed to provide adequate quality for the connection and does not cause any
harm, since increased transmit power is compensated by the fading channel. For
neighbouring cells however this means additional interference. The transmit
power increase (TxPowerlnc) is used to reduce the reuse efficiency according to
equation (9). In equation (4) i should be replaced with term TxPowerlnc x i in
case the mobile station transmit power increase is significant.
Soft handover gain is discussed already in [3]. Handovers - soft or hard provide gain against shadow fading by reducing the required fading margin. Due
to the fact that the slow fading is partly uncorrelated between cells, and by
making handovers the mobile can select a better communication link.
Furthermore, soft handover (macro diversity) gives an additional gain against
fast fading by reducing the required Eb/N0 relative to a single radio link. The
amount of gain is a function of mobile speed, diversity combining algorithm
used in the receiver and channel delay profile. More about SHO gain can be found
in Section II.3 .
II.2 Receiver sensitivity estimation

In the link budget the BS receiver noise density is estimating the noise level over
one W-CDMA - carrier. The required receiver SNR contains the processing gain
and the loss due to the loading. The loading used is the total loading due to
different services on the carrier in question. The required signal power (S)
depends on the SNR requirement, receiver noise figure and bandwidth.
where
where Nf, is the receiver noise figure, is the Boltzmann constant, T0 is the
absolute temperature and is the loading. In some cases the basic
noise/interference level is further corrected with the man-made noise term.
II.3. Shadowing margin and soft handover gain estimation

The next step is to estimate the maximum cell range and cell coverage area in
different environments/regions (see the example in Table I). The RLB is
estimating the maximum allowed isotropic path loss, from that value a slow
fading margin, related to the coverage probability, has to be subtracted. When
estimating the coverage probability the propagation model exponent and the
standard deviation for the log-normal fading must be set. If the indoor case is
considered the indoor loss is from 15 to 18 dB and the standard deviation for lognormal fading margin calculation is set to 10-12 dB. In the case of outdoor
coverage typical standard deviation value is 7 to 8 dB. Typical propagation
constants range from 2.5 to 4. Traditionally, the area coverage probability used
in the RLB is for single cell case [1]. The required probability is 90 to 95% and
typically this leads to 7-8 dB fading margin, depending on the propagation
constant and standard deviation of the log-normal fading. Equation (13)
estimates the area coverage probability for single cell case:
where
Pr is the received level at cell edge, n is the propagation constant, X0 is the

average signal strength threshold and a is the standard deviation of the field
strength.
In real W-CDMA cellular networks the coverage areas of cells overlap and the
mobile station is able to connect to more than just one serving cell. If more than
one cell can be detected the location probability increases and is higher than
determined for a single isolated cell. Analysis performed in [2] indicates that if
the area location probability is reduced from 96% to 90% the number of base
stations is reduced by 38%. This number indicates that the concept of multiserver location probability should be carefully considered. In reality the signals
from two base stations are not completely uncorrelated, and thus the soft
handover gain is slightly less than estimated in [2]. In [3] the theory of the multiserver case with correlated signals is introduced:
where Pout is the outage at the cell edge, SHO is the fading margin in the case of
soft handover, a is the standard deviation of the field strength and for 50%
correlation a = b = 1/ 2. With the theory presented for example in [1] this
probability at the cell edge can be converted to the area probability. In the
W-CDMA link budget the SHO gain is needed. The gain consists of two parts:
combining gain against fast fading and gain against slow fading. The gam
against slow fading is dominating and it is specified as:
If we assume 95% area probability, n = 3.5 and the standard deviation is 7 dB

the gain will be 7.3 dB - 4 dB = 3.3 dB. If the standard deviation is larger and
the probability requirement higher the gain will be more.
II.4. Cell range and cell coverage area estimation

Once the maximum allowed propagation loss in a cell is known, it is easy to
apply any known propagation model for the cell range estimation. The
propagation model should be chosen so that it optimumally describes the
propagation conditions in the area. The restrictions of the model are related to
the distance from the base station, the base station effective antenna height, the
mobile antenna height and the frequency. One typical example for a macro
cellular environment is Okumura-Hata. Equation (16) presents an example of a
Okumura-Hata path loss model for an urban macro cell with base station antenna
height of 25 m, mobile antenna height of 1.5 m and carrier frequency of 1950
MHz [4].
After choosing the cell range the coverage area can be calculated. The
coverage area for one cell in hexagonal configuration can be estimated with:
where S is the coverage area, r is the maximum cell range and K is a constant,
depending on the network topology. The number of sectors is typically from
1 to 3. In the case of W-CDMA reasonable values are up to 6 sectors. In the case
of 6 sectors the estimation of the cell coverage area becomes problematic, since
Table I. Example of a W-CDMA RLB.

DL
UL
Transmitter power
Tx antenna gain
Cable/body loss
Transmitter EIRP
(incl. losses)
Thermal noise density
Receiver noise figure
Receiver noise density
Receiver noise power
Interference margin
Required Ec/Io
125.00
20.97
0.00
2.00
18.97
-174.00
5.00
-169.00
-103.13
-3.01
-17.12
mW
dBm
dBi
dB
dBm
a
b = 10 log10(a)
c
d
e = b + c-d
1372.97
31.38
18.00
2.00
47.38
- 174.00 dBm/Hz
dB
dBm/Hz
dBm
dB
dB
8.00
g
- 166.00
h = f+g
i = 101og10(W)+h -100.13
- 10.09
j
-7.71
k = 10 log10(Eb
/N 0 /(W/R))-j
- 107.85
Required signal power [S] -120.26 l = i + k
0.00
18.00 m
Rx antenna gain
2.00
2.00
n
Cable/body loss
95.00
95.00
Coverage probability
(outdoor requirement)
0.00
0.00
indoor (requirement)
85.62
85.62
Outdoor location
probability (calculated)
32.33
32.33
Indoor location
probability (calculated)
outdoor
outdoor
Limiting environment:
7.00
7.00
Log normal fade
constant outdoor
12.00
12.00
Log normal fade
constant indoor
3.50
3.50
Propagation model
exponent
-7.27
0
-7.27
Log normal fade margin
2.00
0.00
Handover gain
P
(inc. any macro
diversity combining
gain at cell edge)
-5.27
-7.27
Slow fade margin
q =o+p
0.00
r
0.00
Indoor loss
0.00
s
0.00
TPC headroom
(fast fade margin)
147.96
147.96 t = e- l + mAllowed propagation loss
n + q + r- s
dBm
dBi
dB
%
%
%
%
dB
dB
dB
dB
dB
dB
dB
dB
a six-sectored site does not necessary resemble a hexagon. A proposal for the cell
area calculation at this stage is that the equation for the omni case is used also in
the case of 6 sectors and the larger area is due to a higher antenna gain. The more
sectors are used the more careful soft handover overhead has to be analysed to
provide an accurate estimate. In Table II some of the K-values are listed.
Table II. K-values for the site area calculation.

SITE
CONFIG.
K
1 omni
2-sectored
3 -sectored
6-sectored
2.6
1.3
1.95
2.6
II.5. Capacity and coverage analysis in the initial planning

phase
Once the site coverage area is known the site configurations in terms of channel
elements, sectors and carriers, and site density (cell range) are to be selected so
that the supported traffic density can fulfil the requirements. An example
dimensioning case can be seen in Section IV. The W-CDMA RLb is slightly more
complex than the TDMA one. The cell range depends on the number of
simultaneous users (number of channels/users in terms of interference margin,
see equation (6)). Thus the coverage and capacity are connected and in the very
beginning the operator should already have knowledge and vision of the
subscriber distribution and growth since it has a direct impact on the coverage.
Finding the correct configuration for the network so that the traffic requirements
are met and the network cost minimized is not a simple task. The number of
carriers, number of sectors, loading, number of users and the cell range all have
an impact on the result.
III. Detailed planning process

III.l. Introduction to a static radio network planning
simulator
In this study the simulator first introduced in [9] was used. It is of static nature
and needs as inputs a digital map, the network layout and the traffic distribution
in form of a discrete user map. Each of the users can have different terminal
speed and uses a different service (bit rate, activity factor, which both can be
different for uplink and downlink). Therefore each mobile station gets assigned
an individual E b /N 0 requirement imported from link level simulations. The
simulator itself consists of basically three parts - initialization, combined uplink
and downlink analysis and post processing phase. Following the initialization
part, round after round in the main part of the tool, both the uplink and downlink
for all mobile stations are analysed and after the iterations have fulfilled certain
convergence criteria, in the final step, the results of the uplink and downlink
analyses are post processed for various graphical and numerical outputs. On top
of these results, for selected areas (which also can consist of the whole network)
Figure 2. Static simulator overview.
area coverage analyses for UL and DL dedicated channel as well as for common
channels (common pilot CPICH, broadcast control channel BCCH, forward access
and paging channel FACH and PCH on the P-CCPCH and S-CCPCH) can be
performed. In case a second carrier is present in the network area, either used by
the same operator or by another operator adjacent channel interference (ACI) can
be taken into account. Only when the second carrier is assigned to the same
operator, load can be shared according different strategies between the carriers
(IF-HO).
III.2. Initialization phase

In the global initialization phase the network configuration is read in from
parameter files for base stations, mobile stations and the network area. Some
system parameters are set and propagation calculations are performed. In the
following step, requirements coming from the link level simulations are
assigned to base stations and mobile stations. After some initialization tasks for
the iterative analysis - setting default transmit powers and network performance
- the actual simulation can start.
III. 3. Combined uplink and downlink analysis

111.3.1. Uplink iteration step
The target in the uplink iteration is to allocate the mobile stations' transmit
powers so that the (interference+noise)-levels and thus the base station
sensitivity values converge. The average transmit powers of the mobile stations
to each base station are estimated so that they fulfil the base station Eb/N0
requirements. The average mobile stations' transmit powers are based on the
sensitivity level of the base station, service (data rate) and speed of the mobile
station and the link losses to the base stations. They are corrected by taking into
account the activity factor, the soft handover gains and average power raise due
to fast transmit power control. The impact of the uplink loading on the base
station sensitivity (noise rise) is taken into account by adjusting it with (1- ).
can be defined by equation (5).
After the average transmit powers of the mobiles have been estimated they
are compared to the maximum value allowed and mobiles exceeding this limit
are trying IF-HO if allowed or are put to outage. Now the interference analysis can
be performed again and the new loading and base station sensitivities are
calculated until their changes are smaller than specified thresholds. Also when
the uplink loading of a cell exceeds specified limits, mobile stations are moved
to another carrier if allowed (IF-HO). Otherwise they are put to outage.
III.3.2. Downlink iteration step
Similarly to the uplink the goal of the downlink iteration is to assign the base
station transmit powers for each link (including SHO connections) a mobile
station is having until all mobile stations receive their signal with the required
carrier-to-interference-ratio, C/I, defined by equation (18).
where EbNoMS is the received Eb//N0 requirement of the MS depending on

terminal speed and service. The actual received (C/I)m of MS m is calculated
using MRC according to equation (19) by summing the C/I values of all links k,
k = 1, ..., K the mobile station m is having.
where Pk is the total transmit power of the base station to which link k is
established, Lpkm is the link loss from the cell k to the mobile station m, k is the
cell specific orthogonality factor, pkm is the power allocated to the link from base
station k to mobile station m, Ioth, m is the other cell interference and Nm is the
background and receiver noise of MS m.
The initial transmit powers are adjusted iteratively according to the
difference between the achieved and the targeted C/I value until convergence is
achieved. The process requires iteration, since the C/I at each mobile station is
dependent on all the powers allocated to the other mobile stations and it is not
known a priori whether a link can be established or not. In case either certain
link power limits or the total transmit power of a base station is exceeded mobile
stations are performing IF-HO if allowed or taken out randomly from the network.
In a further step for each mobile station it is checked whether the received
Ec/I0 value is above a user defined threshold so that the mobile station can
reliably measure the base station and synchronize to it. Also, here if the threshold
given is exceeded, the mobile station tries IF-HO or is put to outage. A flow chart
for the detailed iteration steps can be seen in Figure 3.
Figure 3.
Flowcharts for the UL and DL iteration steps.
III.3.3. Adjacent channel interference calculations

The adjacent channel interference influence due to the two possible carriers either from own network or from a competitive operator's network in the same
area - is taken into account by filtering this interference with a channel
separation dependent filter. In both directions UL and DL the adjacent carrier
filtering is implemented as a two-fold process. In UL, one filter for the mobile
stations has been implemented indicating that it is out of band radiation
(acpFilteruL). This filter is used to indicate how much power the mobile station
is leaking into the other carrier's receiving band (Adjacent Channel Leakage
Ratio, ACLR). For the base station in the uplink another filter (aciFilterUL) has been
implemented. This filter is indicating the selectivity of the base station's receiver
in multi-carrier situation, i.e. how big a portion of the adjacent channel power is
received by the base station as adjacent channel interference power (Adjacent
Channel Protection, ACP). Also this filter setting depends on the carrier
separation. The adjacent channel interference situation in UL is depicted in
Figure 4. In the simulations these two filters are combined to a single filter by
equation (20):
Figure 4. UL adjacent channel interference situation.
Adjacent channel interference (IACI) is taken into account when calculating

the UL load according to Equation (21).
Advances in
UMTS
technology 161
Also in the downlink a similar type of filtering is introduced as in the uplink.

One filter for the base stations has been implemented indicating the out of band
radiation of the base station (acpFiZterm). This filter is used to indicate how
much power the base station is leaking into the other carriers receiving band
(ACLR). The filter setting depends on the separation between the carriers. For the
mobile station another filter has been implemented (UCiFdterDL). This filter is
indicating the selectivity of the mobile station's receiver in a multi-carrier
situation, i.e. how much of the adjacent channel interference power is received
by the mobile station ( A C P ) . Also this filter setting depends on the carrier
separation. The adjacent channel interference situation in DL is depicted in
Figure 5. In the simulations these two filters are combined to a single filter by
equation (22):
Figure 5.
DL
adjacent channel interference situation.
Adjacent channel interference (IAcI) in DL is taken into account when

calculating the CII of a single link a MS is having according to equation (23).
where variables are as defined after equation (19), Ioth is interference from other
cells on same carrier and I,, is adjacent channel interference.
III.4. Predicting the network coverage

In all estimations of area coverage probability (UL, DL, DCH, CPICH, BCH, FACH,
PCH) it is assumed that the interference situation is stable. This means that a
certain traffic distribution has been assumed and the detailed UL and DL iteration
has been run. A test mobile is then run through all pixels of the interesting area
and all other MSs which could be served are contributing to the interference. The
test mobile is not influencing the interference situation, therefore the other to
own cell interference ratio will not change and also the total transmit power of
the serving BSs is constant.
III.4.1. UL DCH coverage
In UL direction it can now be estimated whether or not this additional mobile
station using a certain bit rate and having a certain Eb/N0 requirement could get
the service in the chosen geographical location, meaning whether, if the
maximum allowed transmit power of the test MS is enough to fulfil the Eb/N0
requirement at the BS receiver. The transmit power needed for the MS is
calculated with equation (24) and compared to the maximum allowed.
Area coverage probability is finally defined as the proportion of the chosen

area where the additional MS really gets the wanted service under that stable
interference situation. The weakness in this approach is that in reality an
additional mobile station would change the interference situation, for example
some other mobiles could go to outage but in the case of low bit rate services
this effect can be neglected.
III.4.2. DL DCH coverage
In the downlink direction the coverage probability calculation is based on the
transmit power limits per radio link. The main focus is to check, pixel-by-pixel,
whether or not there is enough transmit power per link from base stations in the
active set, if there would be a MS in the pixel, using a given service (bit rate) and
having a given speed. Also here it is assumed that the total transmit powers of BSs
do not change from what was left after UL/DL iteration, i.e. it could be assumed that
the influence of the test MS is replaced by the closest original MS. In iteration
however there may or may not have been a MS in the pixel. The coverage
probability then again is defined as the number of pixels from the whole area
where the service was possible.
III. 4.3. CPICH coverage

In radio network planning the CPICH transmit power should be set as small as
possible but ensuring that the best cell and neighbour cells can be measured and
synchronized and the CPICH can be used sufficiently as the phase reference for
all other DL physical channels. Typically this means that 5 - 10% of the total BS
power is used for CPICH. For each pixel in the chosen area the Ec/I0 in that pixel
is calculated according to equation (25).
where P CPiCH is the CPICH power of the best server, Lp is the link loss to the
best server, PtX i, is the total transmit power of BS i, Lpi is the link loss to BS
i, I ACI is adjacent channel interference, N0 is the thermal noise of the default
MS and numBSs is the number of base stations in the network. The achieved
Ec/I0 is then compared to a user given threshold and the CPICH coverage
defined as the ratio of pixels where the threshold is exceeded compared to all
pixels. The weakness of this modelling for the CPICH EC/I0 coverage is that it is
done only for the best server. In an operating network however, all neighbour
cells must also be measured and therefore all neighbour cells' CPICH EC/I0
should be analysed, too. This could be overcome however by e.g. adding a
threshold to the required CPICH Ec/IQ. Descriptions of other static simulators
can be found for example in [18, 19].
IV. Example of dimensioning case and verification of

dimensioning with static simulations
This section is contains an example of the case of the initial planning phase for
a green field macrocellular operator. In this example only one network
evolution phase is considered, in the real radio network planning case the
traffic growth should be more carefully considered. In this work three cases
have been analysed. The first case is 8 kbps speech only, mobile station speeds
3 (50% of MSs) and 120 km/h (50% of MSs). The second case is dimensioning
speech and circuit switched data (64 kbps), speech mobiles moving 50 km/h,
data terminals 3 km/h. The last case is 144 kbps data only, terminals moving
with 20 km/h. More details related to the case can be found in [14]. The study
consisted of two parts. In the first part the operator's macrocellular network is
dimensioned, i.e. the cell range is estimated with given input parameters. This
study is based on the assumption that the traffic and the QoS information are
available from the operator. In the second phase the network is planned for the
estimated site distance (1.5*R) and the W-CDMA analysis is performed for the
radio network. As depicted in Figure 1 this case is concentrating on the first
half of the radio network planning process: the network dimensioning,
network configuration definition and coverage/capacity planning.
In the first phase the dimensioning task is to generate certain traffic and QoS
requirements for the above mentioned three cases and to dimension the cases
accordingly. The radio network was dimensioned for an urban area of
13 X 13 km2 . The propagation model used was Okumura-Hata with an area
correction factor of 0 dB. The traffic and QoS requirements for all the three cases
is shown in Table III.
Table HI. The traffic and QoS requirements and

other dimensioning parameters.
SPEECH
Subscribers
60 000
Average
traffic/subscriber
Blocking
Data activity
Assumed soft capacity
SHO overhead
Maximum uplink
loading
MS/BS transmit power
Common channel
overhead
Antenna gain MS/BS
Cable/body loss
MS speed
0.060 Erl
Std for the log

normal fading
Peak to average
ratio for total BS
power calculation
Average orthogonality
i and iDL
BS antenna height
2%
95%
50%
20%
40%
60%
SPEECH
AND 64
KBPS CS
DATA
60 000,
12 000
0.040 Erl ,
0.025 Erl
144 KBPS
PACKET
DATA
1%
80%
95%/70%
24 000
3 kbps
50%
10%
40%
65%
10%
40%
65%
24/43 dBm
24/43 dBm
24/43 dBm
10%
10%
10%
0/17 dBi
3/0 dB
3km/h
120 km/h
0/17 dBi
3/0 dB
50 km/h,
3 km/h
0/17 dBi
3/0 dB
20 km/h
7 dB
7 dB
7 dB
4 dB
4 dB
4 dB
50%
55%
30 m
50%
55%
35 m
50%
55%
100%
35 m
In the dimensioning process both the capacity and the coverage have to be
considered. The cell range is determined not only by the RLb, but also on the
capacity requirements. From the selected cell range the coverage area for the
site can be estimated by equation (17). Once the site coverage area is known
the supported traffic density can be estimated. The selected site density must
be such that the traffic density requirement and the coverage requirement in
terms of coverage probability are met. In the following table the dimensioning
results are collected. In the example case the cell range has been solely based
on the uplink results. In the uplink direction the target loading as specified by
equation (6) is limiting the number of channel elements per cell. As it can be
seen already in the dimensioning results, the base station transmit power is
actually limiting in all of the cases. In order to meet the capacity requirements
of either cell range, the number of sectors or the number of carriers should be
changed. In this case only one carrier was used.
In the second phase of the work the dimensioning results in terms of number
of cells and the cell range were used in the corresponding static simulator runs.
The items with bold font in Tables III and IV were used as inputs for the
simulation. The site distance used in the static simulator has been estimated
with D = 1.5*R where R is the cell range obtained from the RLB in
dimensioning. In this study the network scenario was based on a regular grid,
and the network area consisted only of urban area type. The static simulator
Table TV. Dimensioning results.
SPEECH
SPEECH
AND 64
KBPS CS
DATA
144 KBPS
PACKET
DATA
Selected cell range
2.08 km
1.98 km
1.52km
Uplink loading
0.32 3 km/h
0.29 120 km/h
0.40 speech
0.18 data
0.65 data
Service limiting
the cell range
Speech 3 km/h
64 kbps data
144 kbps data
Number of channel 30 3 km/h

30 120 km/h
elements per cell
required to meet the
capacity based on UL
42 speech
6 data
855.6 kbps
throughput ~
6 data
channels
Number of channel
elements per cell
supported by DL
21
3 km/h
19 120 km/h
27 speech
6 data
5 data
Required number
of 3-sectored sites
(based on UL)
20 (60 cells)
22 (67 cells)
37 (112 cells)
described in Section 111 was used in this study. The %/No requirements and the
traffic distribution used in the simulation were the same as were used during
dimensioning. In the dimensioning phase only the antenna gain is used. In the
simulations the antenna radiation pattern is also required and an antenna with
65" horizontal beamwidth was selected. The 65" antenna was selected because
it is widely used already in 2G networks with three sectored configurations.
Furthermore, according to [ 111 the 65" provides the optimum performance for
the used network configuration. In all the simulations the antenna gain was
always 17 dBi. The soft handover addition window was set to - 6 dB. In Figure
6 there is an example of the network scenario. The system features used in the
simulations are from [ 101 and the ITU vehicular A channel [5] has been assumed
for the channel delay profile.
According to the simulations in the speech case the dimensioning is too

pessimistic for the uplink capacity. Based on the simulations the network could
support 72 speech users versus the 60 proposed by dimensioning. The main
difference is because of the interference statistics. The other to own cell
interference ratio has relative large standard deviation (0.27, minimum value
only 0.11) and thus there are cells which can carry more traffic than equation (6)
is proposing. With the help of the results in Table V it can be concluded that
current dimensioning is correctly limiting the downlink capacity. According to
the simulations with 65" antenna the downlink should serve 46 users, the
dimensioning is proposing 40. The simulated coverage probability is slightly
higher than the requirement, and thus the site distance in reality could be slightly
increased.
In the mixed traffic case the uplink dimensioning result was proposing 42
channel elements for speech users and 6 for data users to meet the traffic
requirement. The UL simulation result is slightly different from the dimensioning
result. The reason for this is partly that it is very difficult to control the
simulation run so that the ratio of the data users and the speech users would be
exactly as required. The initial requirement was that the data users are 17% of
the total users. After the static simulator run the percentage was 16%. In the case
of UL there are on average 37 speech users instead of 42, but on the contrary
there are 8 data users instead of 6. Therefore one can claim that the uplink
dimensioning result is well in line with the simulation results. The power
amplifier dimensioning for the DL (based on [13]) shows a clear downlink
limitation. The uplink 48 users cannot be served in downlink direction with the
assigned 20 W maximum base station transmit power. When the number of users
in the dimensioning was reduced to 27 speech and 6 data users the dimensioning
estimates the total TX power of 17 W including 10% common channel overhead.
Values used for the power calculation were
iDL (=
"
2_,
Pmi
in
, see [13]) 55% and peak to average ratio for
Lpmi of 4 dB. The downlink is limiting also according to the network

simulations. According to the simulation downlink analysis the network can
support 25 speech users and roughly 5 64 kbps data users simultaneously, the
base station total transmit power being 19 to 20 W, depending on the selected
scenario. In the mixed traffic case the dimensioning of the base station total
transmit power follows extremely well the simulations. In terms of coverage
probability the dimensioning is slightly pessimistic. According to the
simulations the data coverage probability is always better than the required
70%. Since the data service is limiting the cell range the speech coverage
probability is always better than the required 95%.
In the data case the dimensioning was proposing 855.6/144 = 5.94
simultaneous users (assuming 10% soft capacity, and iDL = 55%). This result is
very close to the simulated case, but also in this case the uplink dimensioning is
slightly pessimistic. In the downlink dimensioning with iDL = 55%,
orthogonality 50% and 4 dB peak to average ratio the BS transmit power for the
5 users is roughly 19 W, including common channels (10% overhead) and 40%
soft handover probability. A similar number is also achieved with the static
simulations. In the data case the coverage probability is up to 8% higher than
required. The difference in decibels of coverage probability of 80% and 88% is
2 to 3 dB, depending on the exact parameter values that have been used in the
estimations, see [1]. In coverage limited case this has an impact on the required
number of sites. All simulation results are collected in Table V.
In Figure 7 there is an example of the dominance and coverage analysis plot
for speech services, test mobile speed 50 km/h. In the dimensioning the
requirement used for speech was 95%.
Table V. Simulation results. Only the average results

(averaged over all the cells) are included.
SPEECH
Antenna 65, site distance D = 1.5*R

35
3 km/h
Number of
36 120 km/h
users UL
23
3 km/h
Number of
23 120 km/h
users DL
96.3%
3 km/h
UL coverage
98.8% 120 km/h
probability
0.65
Other to own
cell interference
0.58
Loading
19.34
BS transmit
power/W
SPEECH
AND 64
KBPS CS
DATA
144 KBPS
PACKET
DATA
37 speech
8 data
25 speech
5 data
97.6%
74.4%
0.62
0.75
0.67
18.95
0.62
19.10
87.5%
For the mixed traffic (speech and 64 kbps) case the probabilities are 98%
and 70% respectively. Since data services are limiting the range the actual
coverage probability for speech was higher, i.e. 98%. Similar analysis for the
RT data gave 68% coverage probability. In the dimensioning the used
requirement was 70%.
The results presented here demonstrate that in macrocellular environment the
downlink dimensioning is essential, due to the fact that with the assumptions of
the macrocellular environment and Vehicular A channel conditions the network
is downlink limited.
Uplink dimensioning alone gives too optimistic capacity predictions for the
network. Furthermore, the modelling proposed for the dimensioning is verified
with static network simulations. Comparison of the simulation results and the
dimensioning results shows good agreement.
Figure 7. Example result of the mixed
traffic case. The coverage probability for
speech services.
V. Comparison of a static with a dynamic network

simulator
The scope of this section is to demonstrate that the accuracy of the static
simulator is adequate for radio network planning purposes.
V.1. Introduction to the dynamic simulator

Typically system level simulators operate with the resolution determined by the
feature that changes interference situation most often. In W-CDMA the fast closed
loop power control operating with 1.5 kHz frequency is the algorithm having the
highest frequency, and therefore 1.5 kHz frequency is used in the system
simulator used in the comparison. Conventionally, the information obtained
from the link level tool is linked to the system simulation by using a so-called
average value interface describing the BLER performance by average E b /N 0
requirements. The average value interface is not accurate if there are fast
changes in the interference due to, e.g. high bit rate packet users. This kind of
approach suits well static snap-shot simulations, but cannot be used when
simulating systems with fast power control and high bit rate packet data. With
the dynamic simulator presented however, a so-called actual value interface
(AVI) is used that provides accurate modelling of fast power and high bit rate
packet data [16].
In the dynamic simulator the users are making calls and transmitting data
according to the traffic models. The call generation process for real time
services, such as speech and video, is made according to a Poisson process [15],
[17]. For speech, voice activity and discontinuous transmission have to be
considered. For circuit switched data services, the traffic model is a constant bit
rate model, with 100% of activity.
The calculation of interference is an essential process of the system
simulator. The better the interference modelling is, the more accurate results can
be obtained. The total interference power Ibs(k) received by a base station k is
calculated as follows:
where N is the total number of active mobile stations in the system and m is
index for the observed user. Lpn, k is path loss (attenuation due to distance and
slow fading) between the base station k, and the mobile station n. g/g
multi-path fading normalized to having long term average equal to one and J is
the number of multi-path components. pms(n) is the transmission power of the
mobile n. After the interference calculations, the uplink signal-to-noise ratio
SNRUL can be calculated for the user m connected to the base station k as
where G is the processing gain, ai is amplitude attenuation of path i and J is the

number of allocated RAKE fingers. In (27) it is assumed that the signals received
are combined coherently with maximal ratio combining. In downlink the effect
due to orthogonal codes has to be considered. Due to the multi-path propagation
perfect orthogonality cannot be assumed. For optimal maximal ratio combining,
the downlink signal-to-noise-ratio SNRDL for a user m can be calculated as
where Ims(m) is the total interference power received by the mobile station m, M
is number of base stations in the active set, pbs(m,k) is the transmitting power for
the observed user from the base station k, Pbs(k) is the total power transmitted
from the base station k, ak,,i is amplitude attenuation of the channel tap i and Jk
is the number of allocated RAKE fingers from base station k.
In the dynamic simulator the following items were measured: bad quality
calls, defined as calls having an average frame error rate FER exceeding a
threshold (usually 5% for speech). The minimum call duration is set to 7 sec in
order to increase the confidence of the averaging. Statistical data of these calls
such as coordinates, start and end time and the call duration are recorded .
Dropped calls, i.e. calls that have consecutive frame errors that exceed a
threshold (usually 50 frame errors). Usually dropped calls are considered as
severely poor quality calls. So bad quality and dropped calls can be taken as one
measure whose percentage is referred to the number of started calls after the
warm-up period. Power outage - for speech, this is taken from active terminals
including those that are in DTX. Therefore it is slightly distorted due to the other
half of the users that are in DTX. So the actual outage for terminals that are
actively "talking" is higher, rough value is twice than that of the output. There is
no discrepancy for data. E b /N 0 targets are taken from all active terminals
including those in SHO. So all factors regarding the channel and diversity are
taken into account. Finally a SHO probability histogram of the number of
branches per user was collected.
From the static simulator for all UL and DL connections the histogram of th
transmit powers and their cumulative distribution function are taken. Moreover
the p-th percentiles Qp for 0, 50, 75, 90, 95 and 100% are extracted. Statistics
showing the number and type of SHO connection were gathered. For the whole
simulated area the estimated active set size (AS size) is collected (based on the
CPICH levels.) In each simulated case also the UL loading level was stored.
In the final comparison the total traffic per cell, UL power distribution [dBm],
DL total/link power distribution [dBm], SHO statistics, SHO areas, cell dominan
areas, and not served mobiles (static) versus dropped and bad quality calls
(dynamic) were of interest.
V.2. Comparison results

In this section some of the comparison results are collected. The main
conclusion is that the cell level results (for example loading) are in good
agreement with both the simulation methods. In Figure 8 the number of links is
depicted cell by cell. It can be seen that the number of links per cell follows the
same trend.
Figure 8. Number of links per cell for static simulator and

dynamic simulator cell by cell.
Figure 8 the difference of the cell dominance areas as seen by the two
different simulators is depicted. It can be stated that the difference is minor. In
90-95% of all the pixels both simulators propose the same dominant cell.
The uplink power distribution statistics are collected for the speech and
data case in Table VI. The maximum values do not differ significantly, but
some of the percentile-values are well apart. This could indicate different
power distribution shapes for the two simulators. This cannot be avoided due
to the different nature of the simulators.
Figure 9. Dominance difference in the speech case.

In the downlink direction the transmit power statistics were collected. The
comparison results are collected in Table VII. The difference in downlink
direction has similar trend as in uplink.
The (downlink) link power distributions for the two simulators in speech
case are depicted in Figure 10. The shapes of the distributions are close to each
other. In the case of data the variance is larger.
Table VI. UL Power distribution difference.
MIN
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Q50
Q75
Speech case
-44
-10.38 -1.37
-7
-14.5
-49
-5
-4.12
-5.63
Data case
-41.79 -0.08
7.1
-44
-3
6
-2.21
-2.92
-1.1
Q90
Q95
MAX
5.81
0
-5.81
10.95
4.5
-6.45
20.39
20
-0.39
13.59
15
1.41
15.77
19
3.23
20.03
20
-0.03
Table VII. DL Power distribution difference.

MIN
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Q50
Q75
Speech case
16.12
8.25
13.58
-1
-7
12.5
-9.25
-1.08
-0.38
Data case
18.91 -24.01 24.8
7
25
25.5
0.7
-11.91
0.99
Q90
Q95
MAX
18.37
20
-1.63
18.98
21.5
2.52
24.14
24
-0.14
25.57 25.83
25.7 25.8
0.13 -0.03
26.29
26
-0.29
Another important result is the soft handover behaviour in the simulators. In

Table VIII soft handover statistics from the data simulations have been collected.
In addition to the soft handover overhead the difference in active set sizes
were investigated. These results are depicted in Figure 11.
In a radio network planning phase it is important to identify the outage areas
of the network. In this study the outage prediction of the static radio network
simulator was compared to the result from the dynamic simulator. The main
conclusion of this case is that the problems have a tendency to be distributed
roughly in the same locations for either static tool or dynamic one. The number
of problematic calls cannot be directly compared to each other.
Table VIII. Handover comparison - 64 kbps cs data case.
r
.
HA NDOVER TYPE
SHO
1-way
2-way
3-way
OVERHEAD
Static sim. [%]
83.7
14.9
1.4
17.7
Dyn. sim. [%]
72
23
33
Figure 10. Link power distributions. Left figure for the static simulator
(ave Ueta = average UL loading).
Figure. 11.
AS size difference
in speech case.
VI. Conclusion
In this work the radio network planning process and methods for W-CDMA
networks were introduced and verified. Accuracy of tools is essential to provide
an operator with reasonable information of the required network topology and
hardware requirements. In this work initial planning phase (dimensioning)
methods were introduced. Furthermore the dimensioning results were compared
with results proposed by static simulator. The comparison of the static
simulation results and the dimensioning results show good agreement. In general
it can be stated that the proposed dimensioning methods perform with
reasonable accuracy in cases where the traffic distribution is such that the
network can serve several simultaneous users in each cell. In the case of high bit
rate services the network performance is strongly dependent on the location of
the mobile users and the general interference situation, therefore it can be stated
that also the proposed dimensioning modelling will perform with degraded
accuracy. In the case of high bit rate services with low number of connections
the proposed methods would underestimate the required BS power too much
giving only a long-term average. One way to overcome this problem would be
to use the worst case parameter values (cell edge) for the most demanding
services.
Further, it has been demonstrated that static radio network planning
simulator is giving a realization of the network, which is close to the network
analysed by the fully dynamic simulator. Generally the tools result in a similar
picture of the network. The uplink and downlink power distributions as well as
the cell loading levels and supported links per cell are following the same trends.
Problem zones in the network (dropped/bad quality calls, outage etc.) occur in
the same locations in both analyses. Handover probabilities, active set sizes and
dominance areas are almost the same in both tools. Nevertheless, a static tool is
suitable for network planning. A dynamic simulator however is superior for
benchmarking of radio resource management algorithms and for analysing the
dynamic phenomena in the networks. The reason is not only in the computation
complexity of the dynamic one but also in the fact that the dynamic tool is "not
calibrated", i.e. that the call drop rate is dependent on dropping criteria and thus
the number cannot be taken as exact absolute value. The number could just be
compared with different simulation with the same dropping criterion.
Acknowledgements
The authors would like to thank all their colleagues, especially Mr. Kari Sipil,
Dr. Ari Hmlinen and Dr. Toms Novosad for support and valuable
contributions.
REFERENCES
[1] JAKES (W.C.), Microwave mobile communications, IEEE PRESS,
p. 126.
[2] REUNANEN (J.), "Multiple Server Location Probability in GSM/DCS1800 Cellular
Systems", Master of Science in Engineering Thesis, Helsinki University of
Technology, (1997).
[3] VITERBI (A.J.), CDMA Principles of spread spectrum communication,
Addison-Wesley, p. 198, (1995).
[4] HATA (M.), "Empirical Formula for Propagation Loss in Land Mobile Radio
Services", IEEE Transactions on Vehicular Technology, VT-29, n 3, (Aug. 1980).
[5] Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000,
Recommendation ITU-R M. 1225, (1997).
[6] SIPIL (K.), LAIHO-STEFFENS (J.), JSBERG (M.), WACKER (A.), "Modeling the Impact of
the Fast Power Control on the WCDMA Uplink", Vehicular Technology Conference,
Houston, (May 1999).
[7] SIPIL (K.), JSBERG (M.), LAIHO-STEFFENS (J.), WACKER (A.), "Soft Hand Over Gains in
Fast Power Controlled WCDMA Uplink", Vehicular Technology Conference,
Houston, (May 1999).
[8] TOSKALA (A.), HOLMA (H.), MUSZYNSKI (P.), "ETSI WCDMA for UMTS", Proceedings of
ISSSTA98, South Africa, (Sept. 1998).
[9] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), JSBERG (M.), "Static Simulator for
Studying WCDMA Radio Network Planning Issues", VTC99, Houston.
[10] http://www.3gpp.org/.
[11] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), HEISKA (K.), "The Impact of the Base
Station Sectorisation on WCDMA Radio Network Performance", VTC99
Amsterdam.
[12] A. SAMPATH (A.) et a/., "Erlang Capacity of a Power Controlled Integrated Voice
and Data CDMA System", IEEE 47th Vehicular Technology Conference Proceedings,
3, Phoenix, Arizona, pp. 1557-1561, (May 1997)
[13] SIPIL (K.), HONKASALO (Z.), LAIHO-STEFFENS (J.), WACKER (A.), "Estimation of Capacity
and Required Transmission Power of WCDMA Downlink Based on a Downlink Pole
Equation", in Proc. of the IEEE VTC2000-Spring conf., Tokyo (May 2000).
[14] LAIHO-STEFFENS (J.), SIPIL (J), WACKER (A.), "Verification of 3G Radio Network
Dimensioning Rules with Static Network Simulations", in Proc. of the IEEE
VTC2000-Spring conf., Tokyo, (May 2000).
[15] HMLINEN (S.), HOLMA (H.), SIPIL (K.), "Advanced WCDMA Radio Network
Simulator", Proceedings of PIMRC 1999, Aalborg, Denmark, pp. 509-604,
(Oct. 1997).
[16] HMLINEN (S.), SLANINA (P.), HARTMAN (M.),
LAPPETELINEN (A.), HOLMA (H.),
SALONAHO (O.), "A Novel Interface Between Link and System Level Simulations",
Proceedings of ACTS Summit 1997, Aalborg, Denmark, pp. 509-604, (Oct. 1997).
[17] BRADY (P.T.), "A Model for Generating On-Off Speech Patterns in Two-Way
Conversation", The Bell Systems Technical Journal, 48, n 7, (Sept. 1969).
[18] DEHGHAN (S.), LISTER (D.), OWEN (R.), JONES (P.), "WCDMA capacity and planning
issues", IEE Electronics & Communication Engineering Journal, pp 101-118,
(June 2000).
[19] LABEDZ (G.), LOVE (R.), "A New Time-based Outage Criterion for the Forward and
Reverse Links of DS-CDMA Cellular Systems",VTC98,Toronto, pp. 2182-2186.
Chapter 7
An open software-radio
architecture supporting
advanced 3G+ systems
C. Bonnet, G. Caire, P. Humblet, A. Nordio
D. Nussbaum and R. Knopp
Institut Eurcom, France
A. Enout
UDcast, France
G. Montalbano
Philips Semiconductors, France
T. Hhne
Nokia Research Center, Helsinki, Finland
B. Rimoldi
Mobile Communications Laboratory, Switzerland
I. Introduction and motivation

The presence of several different wireless communication standards and the wide
variety of services provided by mobile communication operators has created the
problem of providing universal seamless connection to customers with different
service requirements at any point on the globe. Software Radio is an enabling
technology for systems aiming at handling several different standards and
different services and thus represents a solution to this problem. Generally,
Software Radio is a very broad term encompassing several levels in the protocol
stack (see e.g. [1, 2, 3] and references therein). Motivated by the worldwide
activity around third generation (3G) mobile communication systems, Eurcom
and EPFL have launched a joint project whose objective is to design and
implement a real-time software radio communication platform to validate
advanced mobile communication signal processing algorithms. The right to
transmit has been granted for one 5 MHz UMTS channel in both France and
Switzerland for experimental purposes.
178 An open software-radio architecture
The platform is characterized by the following major features:

Flexibility, achievable by a software driven system.
Duplex communication.
Multiple antennas transmit and receive signal processing (i.e. joint
spatio-temporal signal processing).
Flexibility remains a key word for a software-defined system. In our case it
serves several purposes. For instance one may perform propagation channel
measurements and transmitter characterization, evaluate the performance of
different signal processing algorithms for both single user and multi-user
systems under different operating conditions. Duplex communication is also
necessary to allow higher-layer protocol testing and services, and to analyse
more complex system aspects, such as multiple-access, power control, and
optimize down-link signal processing from up-link measurements. In a
second phase, the platform will allow Multiple-antenna signal processing, or
more generally, spatio-temporal signal processing (also known as array
processing) since this a very promising ensemble of techniques able to
significantly increase the capacity of wireless communication systems.
The main focus of this paper is on the description of the basic platform
architecture and signal processing to implement the essential physical layer level
procedures of the UMTS standard operating in Time Division Duplex (UMTS/TDD)
mode.
At the physical layer level, Software Radio generally requires the
development of signal processing algorithms suited to implementation on a
general purpose programmable processor (as opposed to analog or digital
dedicated hardware). For this reason the position of Analog to Digital (A/D)
and Digital to Analog (D/A) conversion must be moved as close as possible to
the antennas. The goal is to perform operations like channel selection [4, 5],
synchronization and detection in the all-digital domain, by using high
performance Digital Signal Processors (DSPS). Special care must be dedicated
to the transmission and reception front-end architecture. In fact, this should
be independent of system-dependent parameters like the signal bandwidth
and the symbol (or chip, in a CDMA system) rate. When designing software
radio algorithms it is assumed that a multi-band Radio Frequency (RF) section
takes care of translating the desired signal from a fixed Intermediate
Frequency (IF) carrier to the required RF carrier in the transmitter, and
vice-versa in the receiver. Then we are concerned with the efficient
generation of an IF analog signal from a baseband digital signal (transmitter
front-end) and with the reverse operation (receiver front-end). Once the
end-to-end transmitter receiver architecture is defined the next step consists
in synchronizing the receiver, estimating the channels associated with the
users, and eventually detecting the transmitted symbols. All these operations
are performed in real-time on our platform.
The adopted modulation scheme (BPSK and QPSK), the spread signal
bandwidth (5 MHz), the spreading gain (DS-CDMA with allowed spreading gain
of 1, 2, 4, 8, 16) and the frequency band (around 2.1 GHz) are the same as those
defined in the UMTS/TDD specifications. The TDD mode has been chosen because
under certain circumstances it allows the exploitation of the channel reciprocity
between uplink and downlink in duplex communication, and also reduces the
DSP computational load for a given bandwidth.
The solutions proposed in this paper can also be applied to a wide class
of linearly-modulated digital signals (including most of today's and future
mobile communications standards, like GSM, IS-54, IS-136, IS-95, and DECT
(see e.g. [6] and references therein), UMTS (both FDD and TDD modes) [7],
CDMA2000 (see e.g. [8, 9]) and EDGE (see e.g. [10] and references therein).
II. System architecture

In a first phase, the development of the platform has been based on a single
antenna architecture for the Mobile Terminal (MT) as well as for the Base Station
(BTS). The retained architecture permits to easy enhancement the capabilities of
the platform without redesigning the essential hardware and software parts.
The BTS and the MT are based on the same hardware platform and differ only
in the software implementation. The different hardware parts of the platform are
highly partitioned in order to enable the use of as many standard cards and
components as possible.
The hardware portion of the test-bed consists of 4 elements which are under
software control, namely
1. a PCI bus based reconfigurable data acquisition card (DAQ) based on a Field
Programmable Gate Array (FPGA)
2. an RF front-end
a single stage up/down-conversion from/to a 70 MHz intermediate
frequency (IF) carrier with time-division-duplex (TDD) multiplexing
1 high-speed 12-bit bandpass sampling A/D converter
(fADC = 14.7456 MHz)
1 high-speed 12-bit up-sampling D/A converter
(fDAC= 8 x fADC = 117.9648 MHz)
8 slow D/A converters for controlling various amplifier gains on the RF
card
control for various switches on RF card
3. a clock card for generating sampling clocks (fixed frequency) and local
oscillators (programmable frequency)
We show a simplified block diagram of the entire system for a single antenna
in Figure 1.
We have considered two software implementations, the first using a
combination of commercially available embedded DSP cards and a common PC
and the second using the DSP units (e.g. MMX) of a standard PC under an operating
system proving hard real-time support (e.g. RTLinux). Both run in real-time and
are compatible over the air.
Figure 1. System architecture.
II. 1. RF front-end
A simplified block diagram of our RF front-end is shown in Figure 2. It was
designed in conjunction with STMicroelectronics in Geneva, Switzerland. On
reception (Rx), the RF signal is filtered, amplified and down-converted to a 70
MHz intermediate frequency (IF). The local oscillator frequency is digitally
tunable in steps of 500 Hz. The IF signal is amplified by a digitally tunable gain
control and directly sampled at 14.7456 MSamp/sec. The samples are transferred
via a ribbon cable using high-speed line drivers to the data acquisition unit
(DAQ). The mode of the transceiver (i.e. transmission, reception, calibration) is
fully controllable from the software portion of the platform.
On the transmission end (Tx), samples from the DAQ arrive at a rate of
14.7456 MSamp/sec and drive a hardware up-sampling circuit and a
high-speed D/A converter (117.9648 MSamp/sec) to directly synthesize the 70
Figure 2. RF front-end.
MHz IF signal. This procedure is described in more detail in Section II. 1.2.
This signal is then amplified, up-converted to RF, filtered and amplified by a
variable-gain power amplifier.
Special low-speed lines control the gains of the Rx variable-gain amplifier
and Tx Power Amplifier, LO frequency, as well as the antenna switch. Although
not included in Figure 2, automatic wide-band calibration capabilities (for both
Tx and Rx) are included in the design, which are required for multiantenna
systems (see e.g. [11]). These will be used in a later stage of the project where
the architecture will be extended to implement multiple antenna transceivers.
The basic characteristics of the RF front-end are summarized in Table I.
II.l.l. Passband A/D conversion
Once the RF signal at the antenna has been down-converted to IF, it is sampled
by an A/D converter at a certain rate fADC (Figure 9). Calling r IF (t) the received
IF analog signal and choosing fADC >2W according to [4], because of the
periodicity of the discrete-time signal spectrum, the resulting real sampled signal
r[n] = r I F (n/f A D C ) is pass-band with a spectrum replica centred at fADC/4
(although fIF and fADC at the receiver can be different from fIF and fADC at the
transmitter, for simplicity we use the same notation).
Notice that here in order to avoid signal re-sampling we suppose the rate fADC
to be a multiple integer of the chip rate (i.e. fADC = Ncfc where in our
implementation we set Nc = 4).
Although a base-band version of the received signal can be obtained by
multiplying r[n] by (-j)n and then by low pass filtering, we will show further
how the channel estimation and the data detection processes can be performed
at pass-band.
182 An
oDen software-radio architecture

Table I. RFfront-end characteristics.
Frequency Band
Bandwidth
Transmit Power
(per antenna)
Receiver Sensitivity
Noise Figure
Input lP3
Duplex mode
Rx Gain Control
Tx Power Control
Local Oscillator
RF Calibration
Direct IF sampling
(70 MHz) on Rx
Direct IF sampling
(70 MHz) on Tx
Digital Interface
2 100-2 170 MHz

5 MHz (initially)
1w
-100 dBm
<5dB
> -7 dBm
Time Division
digital tuning, 1 dB steps over 40 dB
digital tuning, 1 dB steps over 80 dB range
digital tuning, steps of a few kHz
in each band
digital control, Tx and Rx
12-bit A/D @ 14.7456 MSamp/sec
12-bit D/A @ 117.9648 MSamp/sec

High speed: Low-Voltage Differential
Signaling (LVDS)
Low speed: 3.3V CMOS line drivers
11.1.2. Passband D/A conversion

converters have an impulse response PDAC
(f) that can be approximated as a
rectangular pulse of duration IIfDAc with frequency response of the form of
sinc(fIfDAc).If the discrete-time input of the D/A converter is the pass-band signal
x[n],then the spectrum of the output signal is given by
D/A
where XV> is the discrete-time Fourier transform of x[n],defined by
Hence, the spectrum replica located at frequency flF is attenuated and

distorted by the D/A impulse response as shown in Figure 8(a). A way to
reduce the attenuation consists of using a DIA converter working at rate
f d = LD,dDAc,
where LDIAis a suitable integer, and up-sampling x[n] by the
factor LDIA.By choosing L,,A such that f d >> flF, the spectrum replica
around fiF falls inside the first lobe of the D/A frequency response since its
first zero is located at fd (see Figure 8(b)). Moreover it is possible to
compensate in part the distortion of the D/A converter by introducing a
pass-band FIR filter between the up-sampler and the D/A converter. The filter
must be designed in order to enhance the spectrum replica at IF while

attenuating the other replicas.
Low complexity implementation. Denote by x"[n] the up-sampled version
of x'[n], given by
and by hD/A[n] the filter impulse response. The filter output is given by
If the filter impulse response has length LD/A, there is only a single
non-zero term in the sum in the right-hand side of the above equation. Then,
x"'[n] = hD/A[m]x'[k], where k= [n/LD/A] and m = n modulo LD/A. Therefore,
the computational cost of the filtering operation consists of one real product
per output sample at rate fd. After the D/A conversion, the continuous-time
signal y(t) is eventually up-converted to the RF carrier and sent to the antenna.
II.2. Data acquisition card

The data acquisition card is a PCI bus-mastering device permitting high-speed
full duplex parallel transfer of digital data from an external device. It contains
the necessary glue logic which connects the input (A/D) and output (D/A) sample
streams as well as some control signals to the main CPU/DSP. A PCI architecture
was adopted since it is the most general purpose bus architecture and is used on
most standard PCS, as well as DSP systems. It consists of two components,
namely a powerful Xilinx Field Programmable Gate Array (FPGA) xcv300 [12]
and a bus-mastering PCI controller PLX9080 [13]. The format is a PCI Mezzanine
Card (PMC) to allow for integration into both embedded DSP architectures and
ordinary PCS. A simplified overview of the DAQ is shown in Figure 3.
Firmware on the FPGA for formatting and transferring data via the DMA
engines of the PLX9080 to the host (DSP, CPU) memory has been developed in
VHDL (Very High Speed Integrated Circuit Hardware Description Language).
After an initial configuration phase, transfers are continuous and completely
transparent to the host, who just "sees" a circular buffer containing samples
acquired from or to be transferred to the external RF front-end.
The basic components of the DAQ are the following:
1. Line Drivers/Receivers for transfer from external devices via
ribbon-cables.
2. A reconfigurable FPGA-based 16-bit bi-directional interface external
device (up to 30 Msamp/sec full-duplex).
Figure 3. Data acquisition unit.
3. A PCI bus-mastering controller for direct transfer of samples to/from

memory (DMA) on PCIbased signal processing units (e.g. DSP cards,
high-performance PCS, workstations (SPARC, PowerPC, Alpha), embedded
processor cards, etc.).
4. A Processor Mezzanine Card (PMC) form-factor for maximum flexibility.
The FPGA is programmed via the PCI bus by serial download.
II.3. Texas Instruments TMSC6201 implementation

The embedded DSP architecture is based on a commercially available dual-DSP
card (Spectrum Signal Processing Daytona [21]). The basic architecture is
shown in Figure 4 and is centred around 2 Texas Instruments TMSC6201
fixed-point DSPS. These DSPS are capable of providing a maximum of 1600 MIPS
each. Our DAQ is placed on the local PCI bus of the DSP board and transfers
samples to/from both of the memory buffers on the DSPS buses. These buffers
are used as temporary storage as the internal (fast) memory of the DSP is rather
small. The DSP DMA engines take care of automatically transferring data to their
internal memory concurrently with the signal processing functions. One DSP is
used exclusively for transmission functions and the other for reception.
The DSPS are used for the front-end processing as described in the previous
sections. Symbol rate data is transferred via the PCI bus to the PC which hosts the
DSP card. This data is processed by the Pentium and handles tasks such as Viterbi
decoding, carrier frequency offset compensation, higher layer protocol stacks,
etc.
Figure 4. Embedded DSP architecture.
II.4. RT Linux-based PC implementation

The second implementation does not rely on embedded DSPS. Here the DAQ is
placed on the master PCI bus of a workstation, possibly multi-processor, running
the hard real-time extension to the Linux operating system, RTLinux [22]. The
software radio runs in kernel space and is integrated into the IPv4 (Internet
Protocol) subsystem of Linux as a network device.
This x86 implementation makes use of the MMX (multi-media extensions)
SIMD (single instruction multiple data) instructions for obtaining maximum
processor efficiency for intensive DSP computations. All DSP routines use
fixed-point arithmetic and are written in C with embedded assembly macros for
time-critical code sections. We typically make use of:
1. MMX packed 16-bit arithmetic (multiply, add, MAC, interleaving, etc.);
2. loop unrolling;
3. software pipelining.
Due to the high-level software structure, this should be portable to other process
or architectures (e.g. PowerPC, Alpha, etc.). For the same reason, it is easily
portable to large-scale SMP (symmetric multi-processing) platforms which could
be useful for advanced base-station implementations.
III. Digital signal processing

This section gives an overview of some theoretical principles on which the
platform software has been implemented. We have implemented a variant of the
UMTS-TDD3GPPstandard [14] and are now implementing a complete subset of
layers 1 and 2 of the true standard (including 1.28 MChips/sec version). The
main difference of the current implementation is that the hardware portion
provides a clock yielding a symbol (chip) rate of 3.6864 Msymbols/sec and not
3.84 Msymbols/sec.
III.1. Frame/slot structure

The frame structure of this TDD implementation is shown in Figure 5. We see that
each frame is composed of 15 slots which can be arbitrarily distributed between
uplink and downlink streams. The first slot in every frame contains the
synchronization sequence and is by default a downlink slot. The synchronization
sequence is used by the mobile terminals to obtain slot timing synchronization.
III.2. Basic transmitter structure

The implementation of the transmitter allows for the generation a composite
signal containing up to 8 variable-rate, variable power data streams per slot. It is
shown in Figure 6. The rates of the different streams is controlled by OVSF
spreading sequences i [n] and the amplitudes by Ai. The choice of i[n] dictates
the spreading factor Li, which ranges from 2", n = 0, ..., 4.
Two possible midambles, m[k], can be inserted having lengths of either 256
or 512 chips. These are superposition of either 3 or 8 cyclic shifts of a training
sequence with a periodic extension. This structure allows for efficient channel
estimation techniques based on the Fast Fourier Transform (FFT).
The beginning of synchronization slots (only BS) contains a primary
synchronization sequence sp [k] of length 256 chips superimposed on the data.
The composite signal is filtered by a 12-tap root-raised cosine FIR filter, p[k],
which simultaneously up-converts the signal to a carrier frequency of /2. To
this end, we choose the sampling rate fDAC according to the classical expression
IF
/l1/4 for a positive integer
Then, we generate the discrete-time real signal
Figure 5. Slot structure.
Figure 6. Basic TX structure.
In this way, the periodic spectrum of x'[n] shows a spectrum replica centred
at fIF (see Figure 7(a)). After D/A conversion, a pass-band filter centred at fIF
removes the other replicas, generating the desired IF modulated signal.
The discrete-time modulation by fDAC/4 in (5) requires a negligible
computational cost since it corresponds to change alternatively the signs of x[n]
as can be noticed in expanding equation (5). In order to avoid aliasing when
taking the real and imaginary part, the sampling rate must satisfy also the
condition fDAC >2W.
III.3. Receiver
In this section we analyse some of the theoretical aspects of the receiver
signal-processing. In particular we give a description of the receiver front-end
architecture shown in Figure 10. Then we focus on channel estimation, matched
filter synthesis and symbol detection.
III.3.1. Frame synchronization
Frame synchronization is achieved using a filter matched to the primary
synchronization sequence to estimate the location of the start of a frame. This is
achieved by filtering the bandpass received signal r[n] as
(a) Spectrum of x[n], x'[n], and y(t) with the integer l = 2 and the sign (+) chosen in (4)
Figure 7. Alternative approach for IF up-conversion.
Figure8. D/A frequency response.
Figure 9. Receiver front-end.
and averaging the rs[n]2 over several frames. The maximum output of this filter
is used to adjust the receive signal strength (via a variable gain IF amplifier) and
synchronization is achieved when the maximum is greater than a pre-defined
threshold. Note that this filtering operation involves purely real quantities. This
is typically the most computationally intensive part of the receiver front-end
since it requires a fairly long filter operating at the sampling rate. The 3GPP
standard uses a hierarchical structure for the primary synchronization sequence
which allows the filter to be implemented as a concatenation of 2 FIR filters of
length 16. A block diagram of the primary synchronization sequence is shown in
Figure 11.
Figure 10. Basic RX structure.
Figure 11. Primary Synchronization Processing.
III.3.2. Channel estimation

Here we consider the training-sequence based multi-user channel estimation
procedure for block-synchronous CDMA described in the UMTS/TDD standard.
In this scheme users are roughly synchronized to a common time-reference
and transmit the training sequence at the same time (user timing errors are
included as an effect of the channel and taken automatically into account by the
estimation procedure). The maximum channel length (including possible timing
errors) is Q symbols and the training sequence sent by each u-th user is a cyclic
shift of the same common base training sequence m = [m[0], m[l],..., m[M- 1]] T
of length M symbols. This solution allows joint estimation of all user channels
if M > QU, where U is the number of interfering users. It is proposed and
described in [16, 17] and with some modifications in [18]. The interested reader
is referred to these papers and references therein for more details.
Under these assumptions we can write the received signal sampled at

frequency fADC = N c f c during the M symbols spanned by the training sequence
as
where
is the received signal,
is a vector containing the channel impulse responses of the U users,
is the u-th user channel filter vector and v is a vector of interference plus noise
samples, assumed to be white. The MNC x MNC matrix A is defined as
where () denotes the Kronecker product and M is a circulant matrix containing

all the possible cyclic shifts (by columns) of the base sequence m. The matrix M
is also circulant and it is unitary similar [19] to the diagonal matrix diag(),
where
and where a is the Discrete Fourier Transform (DFT) of a. After some algebra
[18], it is possible to show that the Least Squares estimation of the overall
channel impulse response g is given by
where DFT and IDFT denote direct and inverse Discrete Fourier Transforms. The
ratio of two vectors should be interpreted as the element-by-element division.
This approach can be applied to both base-band and pass-band signals. The
receiver can also use the a priori information that the signal bandwidth is limited
to W. Notice that this operation in the frequency domain corresponds to low-pass
filtering in the time domain, moreover it reduces the computational cost
since only a part of the MNC products (by the element-wise inverses of in (12))
are computed. Eventually, after the IDFT, the processing gives the estimated
channel complex envelope. The channel estimation procedure is summarized

in Figure 12.
Figure 12. Channel estimation and matched filter synthesis.
III.3.3. Matched filter synthesis and data detection

Given the channel estimates, we are then interested in synthesizing a Matched
Filter (MF) matched to the cascade formed by the user data spreading sequence,
the chip pulse shape filter and the user channel. This is also shown in Figure 12.
The overall impulse response for code i is given by
where Li is again the spreading gain. Using the sampled channel estimate g[k]
we synthesize the discretetime filter fi[k] = fi(t) | t=k/fADC matched to the overall
response as follows
In order to extract the data symbols we filter the received signal with the MF
obtaining
In this setting r[k] is real while fi[k] is complex so the product requires two
real multiplications. To the contrary the baseband samples of r[k] would be at
half the sampling rate but would be complex. So the two complexities are
identical.
Notice that the signal after matched filtering is still pass-band and the symbol
estimates after sub-sampling are given by
where v'i[k] is the complex envelope of the MF output. But since
substituting into (16) and for Nc = 4 we get
In this way the symbol estimates are given by sub-sampling the MF output at
symbol rate without taking care of the demodulation.
III.3.4. Carrier synchronization and decoding
The carrier synchronization is done at symbol rate with a classical
decision directed algorithm [20]. The algorithm then takes a decision on
the symbols and recovers the data (in our example a video stream).
III.3.5. Re-sampling
The baseband processing algorithms such as synchronization, channel
estimation and data detection assume that the signal is sampled with an integer
number, Nc, of samples per chip. In the current implementation we set NC = 4.
This solution avoids utilization of re-sampling techniques at both the
transmitter and receiver front-end. These techniques have been studied in [15]
and will be implemented in the next version of the software.
IV. Validation of the existing platform

The platform described in this paper has been validated by the transmission and
the reception of two user full-duplex real-time video flows in an indoor
environment. Two H263 video streams are transmitted in parallel and decoded
in real time. For this we use the following parameters:
spreading factor 16;
bit rate 397 kbps (peak) per user;
TDD configuration: 1 Tx slot followed by 1 Rx slot (transmission is done
every 2 slots);
two synchronous full-duplex streams per slot;
RF band: 5 MHz at 2.1 GHz.
V. Conclusion
This first demonstration shows that the architecture of the platform is capable
of sustaining real-time communications and is thus promising for future
developments. The platform is currently being enhanced and opened to both
industrial and academic collaboration. The enhancements will consist of:
support for multiple antenna transceivers;
more sophisticated signal processing algorithms;
multi-user detection;
layer 2 (RLC,MAC) functionality.
Collaboration has already begun under the label of the RNRT (Rseau
National de la Recherche en Telecommunications) financing programme
organized by the French Ministry of Industry and Finance. Three projects have
been initiated covering the following topics:
radio sub-system improvement (flexibility and sensitivity);
compliance with the3GPPUMTS/TDD specification;
higher-level protocol stacks;
integration to an IPv6 experimental backbone.
Acknowledgements
The authors would like to thank D. Margairaz and P. Favrat, formerly with
STMicroelectronics, Geneva, for their collaboration in the development of the RF
subsystem. In addition, the help of G. Vaucher and P. Bruchlmeier from
ACORT/EPFL in the fabrication of many of the hardware components was greatly
appreciated.
REFERENCES
[1] SRIKANTESWARA (S.), REED (J.H.), ATHANAS (P.), BOYLE (R.), "A soft radio architecture
for reconfigurable platforms", IEEE Communications Magazine, (Feb. 2000).
[2] "Special issue on software radio", IEEE JSAC, Vol. 4, (April 1999).
[3] "Software radio", IEEE Personnal Communications, Vol. 4, (Aug. 1999).
[4] TSURUMI (H.), SUZUKI (Y.), "Broadband RF stage architecture for
software-defined radio in handheld terminal applications",
IEEE Communications Magazine, (Feb. 2000).
[5] ZANGI (K.), KOILPILLAIR (R.), Software radio issues in cellular base stations",
IEEE JSAC, Vol. 17, pp. 561-573, (April 1999).
[6] RAPPAPORT (T.), Wireless Communications, Englewood Cliffs, (1996).
[7] 3GPP-TSG-RAN-WG1, "TS-25.2xx series", tech. rep., (Jan. 2000).
[8] DAHLMAN (E.), GUDMUNDSON (B.), NILSSON (M.), SKLD (J.), AB (E.R.S.),
"UMTS/IMT-2000 based wideband CDMA", IEEE Communications Magazine,
Vol. 36, pp. 70-80, (Sept. 1998).
[9] "http://www.itu.int/imt/2-rad-devt/index.html".
[10] FURUSKR (A.), MAZUR (S.), MLLER (F.), OLOFSSON (H.), SYSTEMS (E.R.), "EDGE: Enhanced
data rates for GSM and TDMA/136 evolution" IEEE Personnal Communications,
Vol. 6, pp. 56-66, (June 1999).
[11] MOGENSEN (P.), KREDERIKSEN (F.), DAM (H.), OLESEN (K.), LARSEN (S.), "A Hardware
Testbed for Evaluation of Adaptive Antennas in GSM/UMTS", Vol. IEEE PIMRS'96,
pp. 540-544, (Oct. 1996).
[12] XILINX, "Xilinx 2001 FPGA Databook", http://www.xilinx.com, (2001).
[13] PLXTECHNOLOGIES, "PCI9080", http://www.plxtecj.com, (1999).
[14] 3gpp, "3rd Generation Partnership Project", http://www.3gpp.org, (2001).
[15] MONTALBANO (G.), NORDIO (A.), "Front-end software-radio review",
tech. rep., Institut Eurcom, (July 1999).
[16] STEINER (B.), JUNG (P.), "Optimum and suboptimum channel estimation for
the uplink of cdma mobile radio systems with joint detection",
European Transaction on Communications, Vol. 5, pp. 39-49, (Jan.-Feb. 1994).
[17] 3GPP-TSG-RAN-WG1, "TS-25.221 v3.1.0; physical channels and mapping of transport
channels onto physical channels (TDD)", tech. rep., (Jan. 2000).
[18] CAIRE (G.), MITRA (U.), "Structure multiuser channel estimation for blocksynchronous DS/CDMA", Submitted to IEEE Transaction on Communications,
(July 1999).
[19] GOLUB (G.H.), LOAN (C.F.V.), Matrix Computation, The John Hopkins
University Press, (1996).
[20] PROAKIS (J.G.), Digital Communications, NY: McGraw Hill, 2nd ed., (1989).
[21] "Spectrum Signal Processing", http://www.spectrumsignal.com,
[22] FSMLABS, "RTLinux", http://www.rtlinux.org, (2001).
(2001).
Chapter 8
Wireless communications+ + +
R. Steele
Multiple Access Communications Ltd, Southampton, UK
I. Setting the scene

Some five years ago the writer asked groups of experts in mobile radio
communication what would they do after they achieved their cardinal goal of
communications at any time, from anywhere, at any speed. It transpired that, in
general, the question was unwelcome, made them feel uncomfortable, was
considered irrelevant, and for some there was nothing after the great goal had
been realised. That was only five years ago! Now many people have thought and
published treatises on the next phase, and it appears that the notion of multimedia services, particularly Internet Protocol (IP) services, while a user is
nomadic or mobile is what the industry has girded its loins to achieve. So before
we have achieved our first great goal we have modified it, changing our view of
our network from a vehicle to support mainly voice communications, to one that
provides communications and information when and as required with aspirations
of cyberspace. We will be physically in one place at an instant in time, but
electronically we may access remote libraries, participate in virtual businesses,
experience thrills without getting out of our armchairs, and so on. And then
what? To where are we rushing, and will we be happier when we arrive?
This short discourse is concerned with where mobile communications is
heading, and we will return to some of the philosophical issues raised above at
a later stage. Our approach is to start by considering communications for the first
decade of the 21st century. This is not easy as unforeseen profound events are
likely to occur, but we know that third generation (3G) systems will be deployed,
and that during this decade they will evolve. After that we will attempt to move
to later decades. Prediction so far in the future formulated on what is happening
now will be very unrealistic, and instead we will apply preconception based on
how society might be evolving to guess what future networks might be like.
Acknowledging at the outset that we will fail in this objective, we may succeed
in initiating a serious debate on the long-term consequences of our actions now
and in the immediate future.
II. The first decade

In the 1970s and until the first half of the 1990s mobile radio for an engineer new
to the subject was dominated by understanding radio propagation and witnessing
furious debates as to the suitability of different multiple access methods. The
radio interface between users and the network and what was required by the
mobile component of the fixed network became clear. The fixed network needed
special sub-systems, such as databases, switches, gateways, and billing systems,
and inter-working units with the networks of the main telecommunication
companies. Originally most calls were between a mobile user and a PSTN/ISDN
user. The radio interface had a specific multiple access arrangement and a
modified open systems interconnection (OSI) reference model that was
predominantly concerned with the lower three layers of this model. Most calls
on first and second generation networks are voice calls, with some low speed
data offering short message services (SMS). Recently there has been the
introduction of the wireless application protocol (WAP) to provide web-like
services, and the very successful iMODE services in Japan.
By the commencement of the first decade of the 21st century the de facto 2G
world standard, GSM, is beginning to have its Phase 2+ enhancements [1]. There
will be the general packet radio service (GPRS), which will support packet data
rates in excess of 150 kb/sec; high speed circuit-switched data (HSCSD), which
uses TDMA slot aggregation to increase the circuit-switched data rate to 57.6
kb/sec; and the enhanced data rates for GSM evolution (EDGE), which applies
multi-level modulation to GPRS and HSCSD. In addition there is the cordless
telephony system that provides a GSM home base station; a subscriber identity
module (SIM) tool kit that allows standardized access to SIM-based applications;
a standardized intelligent network environment, known as CAMEL; work-group
services including broadcasting and group calls and so on. The consequence is
that the GSM backbone network is required to handle both packet-switched and
circuit-switched data, to operate at much higher speeds and capacity, and to offer
a wide range of services. From a service point of view GSM is becoming a 3G
network.
While this Phase 2+ programme is accelerating, the Europeans decided that
their 3G system, called the universal mobile telecommunications system (UMTS)
would have a radio interface that would be either wideband CDMA for use with
paired spectrum, or a TDMA system with multiple user CDMA signals in a slot
for unpaired spectrum. The wideband CDMA system therefore uses frequency
division duplexing (FDD), while the TDMA/CDMA system employs time division
duplexing (TDD). These UMTS interfaces are radically different from those of GSM
Phase 2+. However, the radio interfaces of both GSM Phase 2+ and UMTS will
198 Wireless communications +++
plug into the same enhanced GSM backbone core network, whose architecture is
displayed in Figure 1. We observe that the conventional GSM network subsystems, such as the BSS, MSC, VLR, HLR, etc, are present, plus elements to support
GPRS. The UMTS terminology for a BSS is a radio network sub-system (RNS), and
for a BS it is a Node B. Figure 1 shows a RNS with interface IUCS for
circuit-switched data from the radio network controller (RNC) (the UMTS version
of a BSC) to the MSC; interface IUPS from the RNC to the packet node SGSN;
interface Iur between RNCS; and interface Iubis between the RNC and the Node B.
It may be expected that with the advent of software radio the BSS and RNC will
be a single entity, as will the BTS and the Node B. We may therefore conclude
that our networks will evolve to handle both packet and circuit-switched data
and with fibre/radio nodes which can adapt to GSM, and UMTS as well as a range
of other radio interfaces. Packet transmissions may eventually be exclusively
preferred over circuit-switched transmissions.
This approach of a core network handling different radio interfaces is not
new. IS-95, now cdmaOne, another global standard, was conceived as a radio
interface, and its original version was a twin of AMPS in a dual-mode mobile. As
is-95 was CDMA and AMPS analogue FM using FDMA, the new digital system had
to work on the IS-41 AMPS backbone network. Nowadays cdmaOne mobiles are
often single mode, operating in both the 800 and 1900 MHz bands. The
evolution of cdmaOne to the 3G system cdma2000 is relatively straightforward.
cdma2000 will be either wideband CDMA spanning a bandwidth of multiples of
1.25 MHz, or it will use multi-carriers, where each carrier has the cdmaOne 1.25
MHz channel occupancy. Thus the lowest state of cdma2000 is cdmaOne (but an
enhanced version).
The expectation of a single global 3G standard has not materialized. Instead
the ITU calls its IMT-2000 a family of 3G systems that will be inter-worked [2, 3].
The terrestrial component has five systems; the two UMTS systems called IMT DS
(direct sequence) and IMT TC (time code); the cdma2000 system called IMT MC
(multi-carrier); IMT sc (single carrier) which is better known as EDGE; and
strangely IMT FT (frequency time) which is the DECT system. There is also the
mobile satellite system (MSS) component to provide global coverage at low
capacity.
Harmonization is in progress to inter-work UMTS and cdma2000, and as EDGE
is on the UMTS road map, then early software radios (which will not include
down-conversion using a DSP) will be able to combine these systems. Further,
MSSS, like Globalstar, are able to inter-work with both GSM and cdmaOne. So the
realization of a single radio interface that can adapt to most cellular systems is
expected during the first decade.
An interesting question relates to operators who have, say, GSM spectrum in

one band for which they pay a modest licence, and 3G spectrum they acquired
at huge cost in a spectrum auction. How will they use these bands when the
mobiles can tune to either band, or perhaps to both bands at the same time, using
software radio? Will they have speech and low speed data on the old 2G bands,
only high bit rate services on the 3G bands, and will they be allowed to acquire
small pieces of spectrum in other bands where they could offer private mobile
radio/special mobile radio (PMR/SMR) services? Will they operate in unlicensed
bands? Will regulators grant them the freedom that they might expect of the
market place? Attempting to accommodate in cellular structures roaming users
having bit rates from a few kilobits per second to 2 Mb/sec with different
quality-of-service (QoS) agreements in one spectrum band is a task of
monumental difficulty. Network design is significantly simplified if similar bit
rate/QoS services are grouped together in different bands. There is another key
question that should be asked on this issue, and that is, should operators be
forced to share bands? By band sharing a wider bandwidth is available and
hence higher bit rate services can be offered. If bands are shared, other than
Figure 1. Network architecture for GSM and UMTS and their evolution.
AUC = Authentication centre
BSS = Base station system
EIR = Equipment identity register
GGSN = Gateway PPRS
support node
HLR = Home location register
RNC = Radio network controller
SGSN = Serving GPRS support node
BSC = Base station and controller

BTS = Base transceiver station
Gb = Interface between BSS and
SGSN
GMSC = Gateway MSC

MSC = Mobile switching centre
RNS - Radio network sub-system
VLR = Visitor location register.
introducing fairness algorithms to make sure an operator is not gaining an unfair

advantage, we need to develop very powerful anti-interference/interference
avoidance algorithms. It is very difficult to predict in this area because of the
current serious consequences affecting the industry caused be auctioning
spectra.
Most commentators and industrialists are betting that IP services to mobiles
is the big growth industry for the first decade. Since they all seem so confident
of their predictions they must have done a lot of market research, or if you are a
sceptic, a lot of navel gazing. During the first decade the bandwidth will limit IP
services, unless a user goes to nodal cells [4], i.e., high bit rate "watering holes".
To achieve widespread use of 2 Mb/sec services, huge amounts of infrastructure
will be needed so that nomadic and stationary users will always be close to a
picocell. For others, IP services will probably operate at 64 kb/sec or 128 kb/sec
(it may be difficult to accommodate roaming 384 kb/sec users), and for such
services to succeed they must be personalized, e.g., personal navigation,
observation of your house for security reasons, monitoring your children at play,
etc. Video is therefore a vital service, but it will be of relatively low quality,
except for those at the "watering holes".
With the advent of body wireless area networks (B-LANS) users will have a
distributed network around their bodies with some radio nodes that will
communicate with the network. The B-LANS are the successor to the hand-held
mobile station, and they will have much higher processing power and memory
storage, and an array of displays. Consequently, non-real time IP services could
use low speed links from the network, store the data in the B-LAN, and display
the information rapidly as if it was a wideband real-time service.
The fibre network was initially used for trunk lines in the fixed network, but
slowly optical fibre has spread over the entire densely populated regions. At the
beginning of the first decade of the 21st century there are few fibre lines to the
home, but the asymmetric digital subscriber line (ADSL) is being used to increase
the capacity of copper wires so that homes can have higher bit rate services.
However, fibre is innately inexpensive with huge capacity, and as the decade
progresses more fibre will be laid directly to offices and homes. Whether it be
ADSL, or preferably fibre, that is used, we may expect the developed countries to
wire-up their urban areas, and in later decades their rural environments. This is
not an over-ambitious task as countries have already been wired for electricity.
As this telecommunications wire-up gathers pace, what is the role for mobile
radio communications? With fixed high capacity links in every building we will
be able to form very small radio cells. The base stations will often be of less than
a matchbox in size, radiating very low power. All of us will live near a base
station and none of us will care! No-one will say they are dangerous, nor
unsightly as they will not be visible, being embedded in building and city
infrastructures. With base stations (BSs) costing very little, they will be deployed
ubiquitously along with cameras, and whether they will be used regularly or not
will not be important. They will be installed for functionality and convenience,
in the same way that lighting and power outlets are used.
Let us pause to consider the implication of myriads of small BSS and their
accompanying MSS radiating low levels of electromagnetic energy over a wide
frequency range and very small coverage areas. The result will be that people
will be exposed to frequent low level electromagnetic fields. This is in contrast
with the current situation of few BSS and MSS transmitting at much higher power.
Given this the expectation is that the average exposure in the future will be less.
Along with the miniaturization and widespread deployment of BSS will go the
installation of physically small switches/routers and databases, although this
may not happen until the second decade. Gone will be the concept of accessing
remote massive switches and databases. Decision making regarding the efficient
use of resources, particularly radio resources, will need to be fast and tailored to
local events. Knowledge of available resources, where the mobiles are currently
located and where they are predicted to be going, QoS agreements relating to the
active mobiles, current link qualities of all links, and so on, will be locally
utilized in deciding what actions to take. Local mobile networks will have
generic similarities with campus networks having private branch exchanges,
although the switches and databases in future mobile networks will be small and
more localized, e.g., a switch/router in an office block or large office. So with
microcells go micro-networks with their small components. Compared to the
present day the equipment effectively disappears in that it will be coffee mug
size.
We mentioned that many BSS will be co-sited with cameras. Tiny in size, the
cameras will be difficult for people to spot their presence, but they will send and
store images of us. While the ubiquitous camera will be an important aid in
decreasing crime, and allow us to observe events at any distance, they could also
be used by governments to ensure that we obey their wishes.
The small radio cells will be used in numerous ways. There will be wireless
local area networks (W-LANS) that will connect home appliances and
entertainment systems and interact with a user's B-LAN. Entertainment will
follow a user around, appliances will be remotely serviced, and complex security
systems will be a growing business. Whenever possible, people will use the fibre
network for accessing the Internet, avoiding the mobile component with its
limited bit rate. Mobile Internet will be for personalized services when nomadic
or on the move. Logging onto the Internet will need to be much faster and
browsing more effective. Users will prefer closed community networks enabling
them to use services, do shopping, etc, with the feeling of security, and with a
quality-of-service they find acceptable. These closed networks will enable
employees to work at home, or elsewhere, creating secure virtual businesses.
III. After the first decade

Predicting events is very difficult, predicting when events will happen is even
more so. Accordingly, we will discuss possible events without saying when
they might occur. Let us start by asking what will happen to our existing
technologies.
III.1. Winners and losers

There are numerous radio systems that have appeared at a time when there was
a social need for them. The social need is like a time window over each
technology, and eventually the technology slips out of the window and is no
more. The first mobile radio system was wireless telegraphy and was much
needed by society. In particular it enabled communication with ships as they
moved around the world. This technology relied on ground stations. It has now
gone, superseded by single sideband (SSB) communication via the ionosphere,
which in turn was replaced by satellite communications.
International voice communications were handled by submarine cables
(point-to-point) and then wireless-voice communications via the ionosphere
enabled any point-to-any point communications. Nowadays the fixed network
spreads across the globe via a mesh of connections, and for inaccessible
locations communications are provided by MSSS.
Television broadcasting had a dramatic effect on peoples' lives, changing
social habits, opinions and behaviour. The transmissions were placed in the
VHFand UHF bands that, at that time, were sparsely occupied. Now these bands
are coveted by mobile operators because they support their cellular
communications. As a consequence, the broadcasting companies will eventually
be forced to deliver their services via fibre and satellite. Broadcast companies
will not be willing to part with their spectrum (and they do think that nature gave
it to them!) and will try to provide value added services by using broadcasting
for high bit rate downlink transmissions, and low bit rate uplink transmissions
using other networks. However, those services that utilize broadcasting channels
can never be as well targeted at individuals as those that can be provided via
mobile networks or by fibre networks, so television broadcasting in the VHF and
UHF bands will be one of our losers. Broadcast services such as sport, news,
weather, will continue, delivered by GEO satellites or fibre networks, but they
will not be in the mobile frequency bands.
Private mobile radio (PMR) has been used by special groups, such as the
emergency services, police, and transport companies, for many years. PMR was
the precursor to cellular radio. It lingers on into the age of cellular with
insufficient spectrum and without the huge market size of cellular that forces
down the costs of mobile stations and network infrastructure. As cellular
becomes more ubiquitous, inexpensive and with a wide range of services, it will
create virtual PMR networks within its cellular architecture. The result will be that
conventional PMR will cease.
Wireless local loop (WLL) delivering point-to-point services to the customer
via wireless from a BS, has a continuing role to play in developing countries
where installing fibre infrastructure can be difficult and/or expensive. In the
developed world, WLL will linger in the USA, but eventually it will be a blip in
history. It will die because of spreading fibre infrastructure and the wide use of
cellular radio.
While satellite systems have a long term role in monitoring and surveillance,
they have only a medium term existence for communications (as distinct from
TV broadcasting). They will be displaced by aerial platform technology and other
space systems as we will discuss later.
Of the systems existing at the beginning of the 21st century the one that will
last the longest is cellular. It is the king of the wireless world. Yet as time goes
on most cellular communications will not be between people, but between
machines.
III.2. Releasing radio spectrum

We will now ponder how much bandwidth we might ideally need for mobile
services. The answer is, never enough. This is the opposite situation facing
optical fibre engineers who ask what will we do with all the capacity we
have.The answer is, wait and we will tell you. But, mobile users must be given
more bandwidth if high bit rate services are to be supported. So what bandwidth
is potentially available? The criterion might be that the transmission frequencies
must be able to significantly propagate around corners.
We note, however, that for high frequencies where diffraction around
buildings is essentially non-existent, then cellular systems can be implemented
[5], but only by the deployment of massive numbers of network nodes. Imagine
a building where every confined space has so many small fixed nodes, i.e. BSS,
transceiving at low power and at frequencies of, say, 60 GHz, that total coverage
is achieved for any mobile node, ie MS. As office and home environments will
have many more fixed nodes than mobile nodes then the utilization of
line-of-sight frequencies becomes an attractive option, as the available
bandwidths are comparatively huge. This technique, of using line-of-sight
transmissions in a cellular mode, will eventually be spread to many outside

locations, e.g., in street canyons.
Returning to the frequencies that have good diffraction and refraction
properties around and within buildings, suppose that 5 GHz of bandwidth
(virtually the first 5 GHz of the radio spectrum) is available. Obtaining this
bandwidth is a political job for the ITU'S World Radiocommunication Conference
(WRC). Suppose in any one location there are five operators, and that each has
1 GHz of bandwidth. Each operator would need to decide how many carriers it
would deploy, and the type of multiple access, modulation, etc, but carrier
spacing of 100 MHz or more would be possible. Now mobile users would have
tens of Msymbols/sec available which might satisfy the most of their needs.
How an operator would acquire such a large bandwidth is a matter of conjecture.
It should not be by auction, which is no more than a government tax on mobile
users, as the amounts of money would be too great. Perhaps we will return to the
sanity of the beauty-contest in allocating radio spectrum.
An alternative way to use a large increase in the amount of spectrum is to
divide it into bands where each band offers a different bit rate service [6]. There
will be a proliferation in the number of service providers (SPS) who will offer
their services on numerous networks. If a SP offers interactive videophone
services, say, that require high bit rates, low delay, and low bit error rate (BER),
then the SP will negotiate with each network operator to ensure that sufficient
infrastructure and functionality exists in their network to satisfy the QoS the SP
will offer to its clients. The SP could also offer subscribers different charges
depending on the QoS, which might mean that different bands are used for the
same service but the grade-of-service (GoS) may be different as one may be far
more congested than the other band. There will also be a yellow pages web site
where a subscriber can go (while mobile) to find a SP that provides the type of
service required. Further, a subscriber will not be limited to one network
operator. A MS on switch-on will scan all the networks, camp on one, find a
suitable SP and get a service. Incoming calls will go to an organization (could be
a network operator) with a HLR to which the user subscribes. Via this HLR,
communication with the VLR of the network on which the user has camped on
will ensue, enabling the user to be paged from this network. Billing will be via
specialist billing agencies.
III.3. Going soft

As software radio interfaces (SRIS) [7] become able to cover a wide frequency
band(s), to reconfigure their transceivers over a packet duration, to house the
necessary IN, and have personalized software; then so will the backbone
network become soft in that it will be a reconfigurable ad hoc network that will
respond locally to local needs, while being able to establish intercontinental

connections. A user with his B-LAN will have a SRI to communicate to the
network fibre/wireless hybrid node (a BS). The B-LAN will be a complete network
distributed about the body, having fibre in the clothing linked to a power
source(s) charged from body energy. This B-LAN will have considerable
processing power and memory. B-LANS will communicate with each other, to
form ad hoc wireless networks, as well as with the numerous network nodes of
home, office, street, MSSS, etc networks. Figure 2 shows a fibre network, nodes
and radio cells.
Figure 2. Fibre network, different radio cells and an ad hoc B-LAN network.
III.4. Air supremacy

The MSS, ICO, Globalstar, and the now defunct (?) Iridium system, were IMT-2000
proposals. They are low earth orbit (LEO) systems, and therefore enable
relatively small hand-held terminals to be used. They are also multi-mode
terminals and can be used on terrestrial networks. 3G terminals will eventually
be able to communicate via all members of the IMT-2000 family of MSS and
terrestrial networks. We may also anticipate that 3G terminals with their soft
radio interfaces will be able to handle signals from broadband satellite networks,
such as Teledesic and Skybridge, as well as from other satellite systems in higher
orbits, although enlarged terminals will be required for non-LEO MSSS.
MSSS do provide global coverage (although the loss of Iridium will make
communications with mobiles in the polar regions difficult), but they also create
large cells on the earth's surface with their multi-beam antennas because they
travel at a height of several thousand kilometres. The high altitude of flight also
results in a substantial delay, and the speed of LEO satellites causes large Doppler
shifts. Their link budgets are such that line-of-sight radio links are preferred,
and communications from within buildings is not, in general, supported.
Furthermore, satellites are expensive, their launch costs are high, and they
cannot be retrieved from space for servicing and technological updating.
All these disadvantages are tolerated because of global coverage.
MSSS will therefore continue to be used until there is a better alternative to
provide coverage and capacity in areas where it is too expensive to deploy
terrestrial cellular systems. Fortunately there are alternative methods. One that
is actively pursued today are high altitude platforms (HAPS). "High" in this
context means the stratosphere, typically 21 km, and the platforms are either
airships or aeroplanes. The altitude selected is where the stratospheric winds are
minimal and planar. At 21 km the HAPS are above the flying lanes of commercial
aircraft, and they are untethered to earth because of aircraft safety and because
there are no suitable cables. An unmanned airship HAP is required by proponents
to stay on-station for about five years, and convert sufficient solar energy to do
this, as well as provide communications. Antenna arrays attached to the
underside of the airship produce multiple beams that form radio cells on the
surface of the earth. Aircraft HAPS can be either manned or unmanned, fly in
tight circles of some 5 to 8 miles in diameter, and carry arrays that form
terrestrial cells.
HAPS have the following advantages. They can be sited over cities to
increase capacity, or in rural areas where they can provide wide area coverage.
Each HAP can produce tens to hundreds of cells and these cells can be tens of
kilometres to a few hundred metres [8-10]. HAPS are linked to ground stations
and to other HAPS to form HAP sky networks. The cells can be instantly moved
by electronically changing the direction of a beam. Compared to MSSS, the
signal delay of HAPS is small, they do not contribute to the Doppler frequency,
the radio link budget is much better, the spectral efficiency is enormously higher
as the cells are small; they can be returned to earth for technology updates; they
are relatively inexpensive to make and launch, one HAP can produce a cellular
network over hundreds of kilometres, and so on.
The aircraft HAP is the HAP likely to be deployed initially, and as a
consequence, the antenna structures cannot be too large. This means they will
operate at relatively high frequencies (e.g., providing local multi-point
distribution service (LMDS)) or if in the 3G bands, they will only create relatively
large cells. The current strategy is for aircraft HAPS to be piloted, and for crews
to do shifts on-station. This means that for one location, three crews and two to
four aircraft are required. Aerial refuelling of aircraft HAPS facilitating long
flights are not proposed. Flight regulations are not an impediment for aircraft
HAPS.
Airship HAPS offer a better long-term solution (at lower latitudes) as they are
some 200 m in length, can have huge antenna arrays and therefore can create
microcells at 3G frequencies. They must, however, have stabilizing systems,
particularly to compensate for tilting of the antenna arrays, and be held onstation as an operator cannot have a network gently floating away!
Other aerial platforms (APS) will be used. Some will be housed on
extraordinarily high structures, where they will create multiple cells that move
around as required. Others will be on tall towers and tethered balloons. Many
civil aviation aircraft could have APS to form terrestrial cells as they go about
their business of delivering passengers to remote destinations. Some APS will be
small and attached to unmanned aircraft that fly to teletraffic hot-spots, etc. APS
can be at any height, subject to aeronautical constraints. The APS will mark the
demise of MSSS. We might also speculate on the longevity of APS, a technology
that has not yet arrived. With the future ubiquity of fibre networks augmented by
radio tails, the effectiveness of AP technology will decline, but it will always
have the feature of providing many cells at switch-on whose characteristics can
be controlled electronically from the ground.
However, communications from space will continue, based on the large
orbiting space stations that were beginning to be assembled at the end of the 20th
century. We may envisage a network of these space stations, descendents of
MSSS, orbiting the world, of huge size, monitoring earth as well as space, and as
a sideline, providing communications [11]. The problems of Doppler shift and
delay associated with MSS would continue, but the spectral efficiency would be
high as the space stations would form very small cells using very large antennas.
This space station system (SSS), with its orbiting space stations interconnected
via laser beams or high frequency radio links, would be able to deliver mobile
services, very high bit rate point-to-multipoint communications, and broadcast
services. They would also interwork with the aerial platform layers and
terrestrial network beneath them.
III.5. From beneath the waves

Numerous optical fibre communication links criss-cross our oceans carrying
vast amounts of traffic. Some of this traffic can be used to communicate with
ships sailing on the oceans and with transatlantic aircraft. To achieve this, sea
platforms can be housed on unmanned surface vessels, powered from the waves,
and connected to the submarine cable beneath the waves by another fibre cable.
By this means a sea platform taps into the submarine cable. A particular
submarine cable will have cables to sea platforms at numerous points along its
length. Antennas connected to a sea platform produce multiple beams, i.e., cells,
for ships in their locality and over-flying aircraft. As the submarine fibres are
laid in many directions the surface platforms connected to them are able to
provide continuous coverage across an ocean. By this arrangement we have
maritime radio cells for ships and aircraft communications connected by a high
speed, high capacity backbone fibre network.
IV. Where is the network heading?

The invention of language was a major step in man's evolution, and the ability
to hold a conversation with a person thousands of miles away was a monumental
achievement. Our first global network was a point-to-point one, then came the
provision of mobility, and later low speed data services. The network has
evolved from an analogue one to a digital one where all signals are transmitted
as a humble bit. Some bits are more important than others, they do not travel at
the same speed, nor do they experience the same delays, nor travel via the same
media. The bit is the atom of communications systems. It can have one of two
voltage levels, and these and its duration are system parameters. It can also be
viewed abstractly as a logical entity; a logical 1 (true) or a logical 0 (false), and
using Boolean algebra we can construct processors, which themselves process
bits. The bit is both carrying information, and in its logical representation, being
manipulated to carry out functions. Scaling this process results in the code
conversion of complex entities, such as the signal from a camera, to a bit stream,
and its distribution to where it is required. It enables us to not only have
communication networks, but to use them to gather and distribute information,
entertainment, and to produce cyberspace. The bit would appear to be here to
stay.
The applications required by our network will become myriad. Our homes
will be micro-networks with wired and wireless LANS monitoring and controlling
how we live, from checking the status of the milk in the refrigerator, to servicing
the washing machine, to presenting pictures of visitors at the front door. We
require to see our house and its contents any time from anywhere. To achieve
this people will use their B-LAN. There will not be a problem with battery power
as our bodies generate some 100 watts and we will harness a fraction of this
power for our body network. This network will have huge processing power and
memory along with holographic and other displays, voice inputs, etc, using
biodegradable electronics. There will be other similar networks in offices, on the
streets, and even in remote areas. Ad hoc networks that are adaptive and
responsive to local demands will be common, but there will be ad hoc network
structures in a hierarchical order having lower order ad hoc networks embedded
within them.
All of these networks, from the international to the one in the refrigerator,
will be soft, i.e., adaptable and flexible to requirements and demands. Such
flexibility, particularly to unforeseen events, requires complex software,
software that is itself changeable. In essence software that, when in a crisis, will
adapt itself. This means code that writes code (we already have this to a limited
extent). These codes will be the software agents (SAS) of the 21st century, the
atom of the software world, the oil that lubricates our interconnecting software
engines.
Let us examine the role of SAs. As an example, consider a B-LAN. When the
wearer speaks, the human interface SA (HI-SA) will, say, visually appear on one
of the user's display systems and ask what duty is requested. Suppose the wearer
is interested in acquiring an antique dressing table. The HI-SA will gather details
for the purchase: the origin of the table, its period, cost, size, when it is needed,
and so on. The HI-SA will have a mobility SA (M-SA) make an access by radio to
a nearby network, visit the yellow pages web site, and obtain a list of service
providers (SPS) who deal in antique dressing tables. The M-SA will then visit the
SAs of the SPs, eventually do a deal, and report the outcome to the user. Suppose
on another occasion the wearer asks the HI-SA what precautions he should take
in climbing a ladder to examine his chimney pot. The HI-SA will ask for
information, and then may need to have a SA devise and use basic equations
involving the theory of statics. If a program is unavailable, the SA will have to
have it written, or ask the M-SA to get either the equations or the result via a SP.
Generally the B-LAN will have the code written locally by body SAs, only
approaching a SP when solving complex problems.
Currently there is much discussion regarding software down-loading to
terminals. Some engineers consider that the terminal should be devoid of much
memory, and consequently the same information will be frequently down-loaded
when required. However, future B-LANS will have formidable storage capacity,
and once acquired, software will be stored locally. This procedure is also in
accordance with people's acquisitive behaviour, e.g. many people prefer to buy
and keep their own books in preference to using a public library.
The network will use SAS to make decisions, and to control, maintain and
adapt their networks. At one extreme we have experienced computer viruses,
which we may think of as malevolent SAS. The SAS created for our networks will
be created by benevolent software robots. They will not, however, be the
equivalent of robots on a car assembly line, programmed to undertake specific
tasks, although they may start off that way. SAS will be large, mobile,
autonomous code and will be created by different engineers with their personal
preferences. They will be conceived to undertake tasks, and inevitably they will
have human-like characteristics. For example, in buying an antique dressing
table the SA will need to be told a lot about the subject as well as handling
business matters. In other words we will need to impart to SAs a lot of human
knowledge, expectations and acceptable goals. Since we also require them to act
independently to unforeseen events that must be dealt with immediately, they
will probably develop genetically. The SAs will therefore occupy a software
world that is psuedo human, and hence SAs will not be good or bad, but have
complex behavioural patterns that will sometimes do things to the networks that
we do not expect.
Although human intelligence evolves extremely slowly, we have been
spectacularly successful in using it to construct modern civilization, and now we
are beginning to understand ourselves. However, we still know relatively little
about our brains and minds. When we consider our network, its processing
power, memory, reconfigurability and evolution are increasing rapidly. When we
apply the word intelligence to the network, or machines in general, it illicits a
howl of protest. This appears to be because intelligence seems to be deemed an
exclusive human attribute, and according to this way of thinking, if a machine
does not have what we think of as human intelligence, it is not intelligent.
You could argue that man has tried to imitate birds, in learning to fly, but has
produced aircraft that cannot flap their wings, dive underwater to catch a fish,
etc. Of course, we did not want to do these things, and we can point out that birds
cannot fly faster than the speed of sound nor carry hundreds of other birds over
thousands of miles. Although man and birds both fly, they do it differently to suit
their respective needs. Similarly, machines and man will both be intelligent, but
in different ways according to their makeup and the goals they need and want to
achieve. So with the passage of time our networks will have highways carrying
enormous quantities of traffic at huge speeds. They will have SAs to allow them
to be self-organizing and reconfigurable. Decisions and executing tasks will be
done at a formidable rate; vast numbers of interconnecting machines will be
accommodated; there will be rising complexity in service provision, and so on.
The SAS will create a software world that will be intelligent.
If this is a glimpse of the future, it follows that we may at this time be
launching a global brain of sorts. This is not like the brain of an individual. It is
more akin to a complex society of interconnecting happenings on many levels;
a representation of our world but in the software domain, the so-called
cyberspace. The hardware of our network, with its multiplexity of wireless and
optical nodes, will somewhat resemble the network's brain, while the SAS will
give it the societal mind.
Industrial strikes by key workers, or picketing of key installations, such as oil
refineries, can bring our societies to a halt in a matter of days. People are unable
to drive their cars to work or to the shops, staff and patients cannot get to their
hospitals, food distribution ceases, and so on. Laws that were fashioned for the
19th and for most of the 20th century seem inadequate to handle the types of
crises that can now suddenly erupt. These flash crises that can bring a country to
a standstill in a few days will be trivial and slow to occur compared to what may
happen in the future when society will be dependent on trillions of
interconnecting processors, SAs, complex ad hoc networks, applications running
at high speed, and so on. If some SAs are forced to make rapid decisions in
response to some unforeseen events, and their decisions result in an almost
instantaneous breakdown of the network, the equivalent of a brain seizure
ensues. Society that will now be so dependent on the brain for almost every type
of decision making, production, distribution and control, will be in a catastrophic
state. In the same way that bulkheads compartmentalize ships so that in the event
of an accident only some parts of the ship will be flooded, we need to design our
networks with software bulkheads. This task should be a major research activity.
A serious fault in the network needs to be rapidly isolated, rather like we isolate
a person with an infectious disease, and dealt with. The "brain" that will make
life for most citizens easy and enjoyable, must be self-healing.
But, will we be able to design and control our global brain with the passage
of time and its growing complexity?
Mankind is faced with a dilemma. As our science is becoming increasingly
complex as time goes by we have fewer people able to understand it. For
example, what proportion of people will be able to comprehend the leading edge
of computer science in fifty years? Very few. Hence we may be inclined either
to do little and arrest mankind's full-ahead mentality, or we go full-ahead
knowing that we risk losing control of our increasingly complex civilization.
There is, however, a way out. Full-ahead to the global brain, but being able to
control it because we biologically change ourselves. We institute changes to
rapidly evolve our own intelligence to keep ahead of our machines. This
involves knowing ourselves in a detailed manner, making biological changes,
and introducing internal body processors so that we can make rapid and complex
decisions. Homo sapiens then forms the genius society and the machines remain
our servants.
Oops, we have gone from 3G to science fiction, or have we?
Acknowledgment
The author thanks the organizers of the ECWT 2000 Conference in Paris, 5-6
October 2000, for permission to amend the paper he presented there, entitled
Full ahead to where?, and to publish it here.
REFERENCES
[1] ZVONAR (Z), JUNG (P.), KAMMERLANDER (K.), "GSM - Evolution Towards Third
Generation Systems", Kluwer Academic Publishers, Boston, (1999).
[2] IEE Conference on 3G Mobile Communication Technologies, London,
(27-29 March 2001).
[3] STEELE (R.), LEE (C.C.), GOULD (P.), "GSM, cdmaOne and 3G systems", John Wiley,
(2001).
[4] STEELE (R.), WILLIAMS (J.E.B.), "Third generation PCN and the intelligent multimode
mobile portable', IEEE Electronics & Communications Eng. J., pp. 147-15,
(June 1993).
[5] STEELE (R.), "Towards a high capacity digital cellular mobile radio system",
lEE Proc. Pt.5, pp. 405-415, (Aug. 1985).
[6] STEELE (R.), "Mobile access technologies - what's next", ACTS Mobile
Communications Summit, Spain, pp. 21-24, (Nov. 27-29, 1996).
[7] HENTSCHEL (T.), HENKER (M.), FETTWEIS (G.), "The digital front-end of software radio
terminals", IEEE Personal Communications, 6, n 4, pp. 40-46, (Aug. 1999).
[8] EL-JABU (B.), STEELE (R.), "Aerial platforms: a promising means of 3G
communications", IEEE Vehicular Technology Conf., Houston, (May 1999).
[9] COLELLA (N.J.), MARTIN (J.N.), AKYILDIZ (I.F.), "The HALO Network",
IEEE Communications, 38, n 6, pp. 142-148, (June 2000).
[10] HASE (Y.), MIVRA (R.), OHMORI (S.), "A novel broadband all-wireless access network
using stratospheric platforms", IEEE vTc '98, pp. 1191-1194.
[11] STEELE (R.), "A vision of wireless communications in the 21ST century",
Shafi (M.), Mizusawa (J.), Ogose (S.), Hattori (T.), "Wireless Communications in
the 21st century', IEEE PRESS, (2000).
Index
3GPP 5
radio interface 6
AAAD 48etseq
AAA-TD 48
access network architecture 8
acquisition indication channel 15
A/D converter 181
adjacent channel interference 160
aliased cross-channels 105
spectral matrix of 105
aliased Tx/Rx global channels 101
asynchronous cell sites 25
three-step search method 25
azimuth-delay 131
spread function 131
BCH (broadcast channel) 21

BER performance 86
CAAAD 48
receiver 48
CAAADIAAA-TD 48
capacity 156
analysis 156
carrier synchronization 193
decoding, and 193
re-sampling 193
CDMA 61 et seq, 98
interpretations 98
models of 98
multicarrier techniques 61 et seq
wideband wireless networks 61 et
seq
cell 154
coverage area estimation 154
range 154
channel 128, 178, 190
estimation 190
flat fading 128
model 128
selection 178
spatio-temporal 129
channelization code 16
code power
interference signal 136
received signal 136
coherent rake combining 30
rake time diversity 30
common pilot channel 14
COMSIC 41
receiver 41
concept 3 (multi-tone - CDMA) 68
coverage 156
analysis 156
CPCH (common packet channel) 22
CPICH coverage 163
D/A converter 182

data acquisition card 183
data detection 192
DCH (dedicated channel) 21
DL coverage 162
UL coverage 162
decoding 193
dedicated physical channel 14
detection 80 et seq
linear 108
detectors 82
without channel coding 82
DIT (discrete Fourier transform) 100
digital signal processing 186 et seq
discrete equalizers (Ts) 98
bank of 98
diversity 30 et seq
DL 162
DCH coverage 162
downlink 158
analysis 158
iteration step 158
TD-CDMA, multisensor extensions
92 et seq
DSCH (down link shared channel) 21
duplex communication 178
dynamic simulator 169
214 Index
environments 116
indoor 116
vehicular 116
equal gain combining
technique (EGC) 74
equalizers, ZF and MMSE 109
ETSI 5 , 7
radio interface 7
standardization 5
FACH (forward access channel) 21

fast TFT 28
SIR measurement-based 28
filter bank 96
channel 96
code-matched 96
flat fading channel 128
forward link 36
frame 186
/slot structure 186
synchronization 188
full load system performance 82
general architecture 9
UMTS 9
global minimum mean square error
(GMMSE)
detection technique 79 et seq
GSM to UMTS, migration 2
IMT-2000 spectrum 4
indication channel 15
acquisition 15
page I5
indoor 116, 121
with fading 121
interference cancellation 77
parallel 77
successive 77
ITU 3
standardization 3
joint linear detection 96
layer I 19
data format
linear detection
joint 96
linear reception
link 35, 36
forward 36
reverse 35
load system 82
half 86
performance
19
96, 108
92 et seq
82
matched filter (MF) 108, 192

data detection 192
synthesis 192
matrix models 103
equivalent 103
maximum ratio combining (MRC) 73
technique 73
MC-CDMA 63,70
receiver 70
systems 70
single-user detection techniques 70
transmitter 70
MC-DS-CDMA 66
minimum mean square error
technique (MMSE) 74
MMSE 109
equalizers 109, 1 12
mobile multimedia
services I
terminals 1
MT-CDMA 68
multi-band transmission 99
multi-carrier spread spectrum 63 et seq
CDMA techniques 61 et seq
multiple-antenna 178
signal processing 178
multiplexing 22
rate matching 22
multi-sensor 92, 114
extensions 92, 114
reception 114
extensions 114
multi-tone-CDMA (MT-CDMA) 68
concept3 68
Index 215
multi-user detection 76, 84
channel coding 84
techniques 76
multi-user transmission 93
network 8, 162
coverage 162
infrastructures 8 et seq
OFDM (orthogonal frequency division

multiplexing) 100
open service architecture (OSA) VII, 10
open software radio architecture 177 et
seq
orthogonal frequency division
multiplexing 100
orthogonality restoring combining
(ORC) 74
page indication channel 15

parallel interference cancellation 77
passband 181
A/D conversion 181
DIA conversion 182
PCH (paging channel) 21
physical channel 13, 14, 21, 22
common packet 22
dedicated 21
down link shared 21
primary-common control 14
random access 14
secondary-common control 14
W-CDMA 13
RACH (random access channel) 21

radio access network architecture
(UTRAN) 9
UMTS 9
radio interface 6 et seq, 7, 203
3GPP 7
ETSI 7
spectrum 203
radio link budget 148
rake 30, 127
receiver structures 127
time diversity 30
random channels 117
rate matching 22
multiplexing 22
receiver sensitivity 152
reception 101
re-sampling 193
reverse link 35
RF front-end 180
RT Linux-based PC
implementation 185
scrambling code 17
secondary-common control physical
channel 14
shadowing margin 153
signal 131, 186
digital processing 186
model 131
signal code power 136
interference 136
received 136
signal to interference and noise ratio
(SINR) 108
simulation results 80
detection schemes 80
simulator, dynamic and static 169 et seq
single-user detection 70
analytical results 75
simulation results 75
systems 70
techniques 70,75
SIR 28
measurement 28
fast TPC based 28
site diversity (softkofter handover) 34
smart-antenna 126, 127
soft handover gain estimation 153
softkofter handover 34
software-radio 177 et seq
architecture 177
open 177
space-time 126
processing 133
results 141
UMTS 126
space-time UMTS uplink processing
127 et seq
216 Index
spatial processing 133

spatio-temporal channel model 129
spectral matrix 105
aliased cross-channels 105
spreading 18
spreading code assignment 16
spreading code assignment, W-CDMA 13
static radio network
planning simulator 156
static simulations 163, 169
dynamic network with 169
successive interference cancellation 77
symbol-band model 105, 106
transmission of 106
synchronization 15
channel 15
code 18
system 130, 137, 179
architecture 179
load definition 137
model 130
T
TD-CDMA 92
downlink 92
multi-sensor extensions 92
TDD (time division duplex) 92
temporal processing 135
Texas Instruments TMSC6201 184
third generation mobile systems,
UMTS/IMT-2000 1 et seq
standardization 3
third generation partnership project
(3GPP) 5, 7, 12
radio interface 7
standardization 5
three-step search method 25
asynchronous cell sites 25
time division duplex (TDD) 92
time interleaving 85
TPC, SIR measurement-based 28
transmission 99
multi-band 99
transmit diversity 39
transport channel 19,21
multiplexing 19
turbo-code, half load system for 86
U
UL, DCH coverage 162
universal mobile telecommunication
system (UMTS) VII, 2, 9, 126
general architecture 9
and GSM 2
radio access network architecture
(UTRAN) 9
smart-antenna space-time 126
uplink processing 126 et seq
uplink 158
analysis 158
iteration step 158
V
vehicular 116, 119, 120
with fading 120
without fading 119
virtual home environment (VHE) 10
W
W-CDMA 12 et seq, 13, 41, 146, 148 et
seq
capacity enhanced technologies 41
improvements 12
physical channel 13
radio link budget 148
radio network planning process 146
et seq
spreading code assignment 13
wide-band model 104
wide-band wireless networks 61
future of 61
wireless 196
communications 196
Z
ZF 109
equalizers 109, 110

Umts PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Umts PDF

Uploaded by

Copyright:

Available Formats

Advances in

This page intentionally left blank

INNOUATIUE TECHNOLOGY SERIES

London and Sterling, VA

First published in 2001 by Hermes Science Publications, Paris

22883 Quicksilver Drive

Hermes Science Publications and GET, 2001

British Library Cataloguing-in-Publication Data

Library of Congress Cataloging-in-Publication Data

Typeset/Design by Jeff Carter, London

2. Improvements in W-CDMA: principles and experimental results

3. Multicarrier CDMA techniques for future wideband wireless

4. Interpretations and performances of linear reception in downlink

5. Smart-antenna space-time UMTS uplink processing for system

6. Radio network planning process and methods for W-CDMA

7. An open software-radio architecture supporting advanced

This page intentionally left blank

The first contribution, "Third generation mobile systems UMTS/IMT-2000" by

3G+ Systems" by C. Bonnet, G. Caire, A. Enout, P. Humblet, G. Montalbano,

FTW, Forschungszentrum Telekommunikation Wien,

This page intentionally left blank

Third generation mobile systems

I.1 Support of mobile multimedia services

Figure 1. Mobile multimedia terminals for UMTS.

Third generation mobile systems

I.2. UMTS: a global mobile system

Advances in UMTS technology

II. International research and standardization

II.2. The standardization of third generation mobile systems

Third generation mobile systems

systems. There could be no global mobile system without a common spectrum

Figure 2. IMT-2000 spectrum.

sectors of ITU: standardization in ITU-T, Radiocommunications in ITU-R. It should

Advances in UMTS technology

uwc-136 (evolution of the American solution ANSI-136 or D-AMPS); this

Third generation mobile systems

III. Radio Interface

Advances in UMTS technology

Table I. Performance requirements for UMTS.

Non-real time services

III.2. The radio interface chosen by ETSI and developed by 3GPP

Third generation mobile systems

Table II. Main characteristics of TDD and FDD modes.

(Time domain duplex)

IV. Network infrastructures

IV.2. Access network architecture

Advances in UMTS technology

Figure 3. General architecture of the UMTS (release 99).

This hierarchical architecture, in which an entity controls several entities at a

Figure 4. UMTS radio access network architecture (UTRAN).

Third generation mobile systems

V. Service principles in UMTS

Advances in UMTS technology

Advances in UMTS technology

II. Physical channel and spreading code assignment

Advances in UMTS technology

In future applications of smart antennas for spot beam transmission, the

Repetition period = Data symbol period

Repetition period = 10 msec frame

User identification (4-512 chips)

Cell (Sector) identification (38,400 chips)

II.2 Spreading code assignment [6]

II.2.1. Channelization code